c-p - y .,.';2 -..p system dynamic analysis of a ......bridget, joseph and blaine, for their...
TRANSCRIPT
NASA-TM-II26_O• d w ,\
7
- ) / y/ - __'>
'C-P_-_"y .,.';2""_-..P
SYSTEM DYNAMIC ANALYSIS OF A WIND TUNNEL MODEL WITH
APPLICATIONS TO IMPROVE AERODYNAMIC DATA QUALITY
A Dissertation submitted to the
Division of Research and Advanced Studies
of the University of Cincinnati
in partial fulfillment of the
requiremems for the degree of
DOCTOR OF PHILOSOPHY
in the Department of Mechanical,
Industrial, and Nuclear Engineering
of the College of Engineering
1997
by
Ralph David Buehrle
B.S.M.E., University of Akron 1985
M.S.M.E., University of Cincinnati 1988
Committee Chair: Randall J. Allemang, Ph.D.
https://ntrs.nasa.gov/search.jsp?R=19970015640 2020-07-30T15:53:42+00:00Z
SYSTEM DYNAMIC ANALYSIS OF A WIND TUNNEL MODEL WITH
APPLICATIONS TO IMPROVE AERODYNAMIC DATA QUALITY
ABSTRACT
The research investigates the effect of wind tunnel model system dynamics on measured
aerodynamic data. During wind tunnel tests designed to obtain lift and drag data, the
required aerodynamic measurements are the steady-state balance forces and moments,
pressures, and model attitude. However, the wind tunnel model system can be subjected
to unsteady aerodynamic and inertial loads which result in oscillatory translations and
angular rotations. The steady-state force balance and inertial model attitude
measurements are obtained by filtering and averaging data taken during conditions of
high model vibrations. The main goals of this research are to characterize the effects of
model system dynamics on the measured steady-state aerodynamic data and develop a
correction technique to compensate for dynamically induced errors. Equations of motion
are formulated for the dynamic response of the model system subjected to arbitrary
aerodynamic and inertial inputs. The resulting modal model is examined to study the
effects of the model system dynamic response on the aerodynamic data. In particular, the
equations of motion are used to describe the effect of dynamics on the inertial model
attitude, or angle of attack, measurement system that is used routinely at the NASA
Langley Research Center and other wind tunnel facilities throughout the world. This
activity was prompted by the inertial model attitude sensor response observed during high
levels of model vibration while testing in the National Transonic Facility at the NASA
Langley ResearchCenter.The inertial attitudesensorcannotdistinguishbetweenthe
gravitationalaccelerationandcentrifugalaccelerationsassociatedwith wind tunnelmodel
systemvibration,whichresultsin a modelattitudemeasurementbiaserror. Biaserrors
overanorderof magnitudegreaterthantherequireddeviceaccuracywerefound in the
inertial model attitudemeasurementsduring dynamictesting of two model systems.
Basedon a theoreticalmodalapproach,a methodusingmeasuredvibrationamplitudes
and measuredor calculatedmodalcharacteristicsof the model systemis developedto
correct for dynamicbias errors in the model attitudemeasurements.The correction
methodis verifiedthroughdynamicresponsetestsontwo modelsystemsandactualwind
tunneltestdata.
ACKNOWLEDGMENTS
I wish to thank my adviser, Dr. Randall J. Allemang, and my graduate review committee,
Dr. Dave Brown, and Dr. Robert Rost, for their guidance and support. ! am especially
grateful to Dr. Clarence P. Young, Jr. for his technical review of this document and his
advice during this research program. I thank my NASA supervisors, Mr. Richard A. Foss,
Dr. William F. Hunter, Dr. William S. Lassiter, Mr. Melvin H. Lucy, and Mr. William F.
Fernald, for their encouragement during my Ph.D. studies. ! would also like to thank Mrs.
Genevieve Dixon of the NASA Langley Research Center for assistance in the finite
element modeling area. Finally, I would like to thank my wife, Barbara, and children,
Bridget, Joseph and Blaine, for their patience and understanding during my studies. This
work was completed under the NASA research program, RTR 274-00-95-01, entitled
"Modal Correction for AOA Bias Errors".
TABLE OF CONTENTS
1.0 INTRODUCTION
1.1 Introduction
1.2 Problem Description
1.3 Literature Review
1.4 Solution Approach
2.0 EFFECTS OF MODEL DYNAMICS ON AERODYNAMIC DATA
2.1 Introduction
2.2 Force Balance Measurements
2.3 Transformation of Balance Forces
3.0 THEORETICAL FORMULATION
3.1 Introduction
3.2 Dynamic Equations of Motion
3.2.1 Lagrange's Equations
3.2.2 Kinetic Energy
3.2.3 Potential Energy
3.2.4 Energy Dissipation Function
3.2.5 Generalized Forces
3.2.6 Equations of Motion
3.3 Modal Analysis
3.4 Simplified Model
3.4.1 Two Degree of Freedom Example
I
2
11
16
18
19
19
25
25
28
28
29
30
30
31
31
34
35
4.0
4.1
4.2
4.3
2.4.2Extensionto MultipleDegreeof FreedomSystem
MODEL ATTITUDE BIASERRORCORRECTION
Introduction
ModalCorrectionTheory
ModalCorrectionImplementation
5.0 EXPERIMENTALVERIFICATION
5.1 Introduction
5.2 Wind-Off DynamicResponseTests
5.2.1 TestSetupandProcedure
5.2.2 CommercialTransportModelTestResults
5.2.3 High SpeedTransportModelTestResults
5.3 High SpeedTransportModelWindTunnelTests
5.3.1 TestSetupinWind Tunnel
5.3.2 DynamicResponseTestsin WindTunnel
5.3.3 WindTunnelTestResults
6.0 CONCLUDINGREMARKS
7.0 REFERENCES
APPENDIXA: Effectof AerodynamicForcesonModalCharacteristics
42
44
45
55
59
59
59
64
72
84
84
86
90
98
102
105
Figure 1.1
Figure1.2
Figure1.3
Figure1.4
Figure2.1
Figure2.2
Figure3.1
Figure3.2
Figure3.3
Figure3.4
Figure3.5
Figure4.1
Figure4.2
Figure4.3
Figure4.4
Figure5.1
Figure5.2
Figure5.3
Figure5.4
Figure5.5
Figure5.6
LIST OF FIGURES
National Transonic Facility model support system.
Schematic of wind tunnel model system.
Effect of vibration on inertial model attitude measurement.
Wind tunnel model instrumentation cavity.
Aerodynamic forces and model coordinate axes.
Influence of angle of attack error on drag coefficient for Cta=0.05.
Reference coordinate systems.
Sting bending in yaw plane, 9.0 Hz vibration mode.
Sting bending in pitch plane, 9.2 Hz vibration mode.
Two degree of freedom model.
Mode shapes for two degree of freedom example.
Harmonic motion of model at natural frequency of 0_y.
Yaw plane mode of model system.
Pitch plane mode of model system.
Flowchart of modal correction method.
Test setup in model assembly bay.
Shaker attachment for excitation in the yaw plane.
Sting bending in yaw plane, 10.3 Hz vibration mode.
Model yawing on balance, 14.4 Hz vibration mode.
Measured mean AOA, estimated bias, and corrected mean AOA
versus yaw moment for sinusoidal input at 10.3 Hz.
Measured mean AOA, estimated bias, and corrected mean AOA
versus yaw moment for sinusoidal input at 14.4 Hz.
4
5
9
10
20
23
26
36
37
38
41
47
53
54
57
6O
62
65
66
68
69
iii
Figure5.7 MeasuredmeanAOA, estimatedbias,andcorrectedmeanAOAversuspitchmomentfor sinusoidalinput at 11.2Hz.
Figure5.8 MeasuredmeanAOA, estimatedbias,andcorrectedmeanAOAversuspitchmomentfor sinusoidalinput at 16.2Hz.
Figure5.9 Stingbendingin yawplane,9.0Hz vibrationmode.
Figure5.10 Modelyawingonbalance,29.8Hz vibrationmode.
Figure5.11 InertialAOA measurement,yawacceleration,andyawmomentversustimefor 9.0Hz sinusoidalinputin yawplane.
Figure5.12 InertialAOA measurement,yawacceleration,andyawmomentversustimefor 29.8Hzsinusoidalinputin yawplane.
Figure5.13 MeasuredmeanAOA, estimatedbias,andcorrectedmeanAOAversusyawmomentfor sinusoidalinputat 9.0Hz.
Figure5.14 MeasuredmeanAOA, estimatedbias,andcorrectedmeanAOAversuspitchmomentfor sinusoidalinput at9.2Hz.
Figure5.15 (Top)MeasuredAOA andestimatedbiaserrorfor 9.2Hz sinusoidalexcitationin pitchwith 0.25Hzmodulation.(Bottom) Correspondingmeasuredbalancepitchmoment.
Figure5.16 (Top)MeasuredAOA andestimatedbiaserror for randomexcitationin pitch. (Bottom)Correspondingmeasuredbalancepitchmoment.
Figure5.17 Measuredandcorrectedmeanangle-ofattackfor sinusoidalexcitationat7.3Hz.
Figure5.18 (Top)MeasuredandcorrectedAOA for sinusoidalexcitationat7.3Hz with0.5Hz modulation.(Bottom)Correspondingbalanceyawmoment.
Figure5.19 Angle-of-Attack(AOA) for first sixty-foursecondsof awind tunneltestonahighspeedtransportmodel.
Figure5.20 (Top)Timedomainresponseof theAOA measuredwith theservo-accelerometersensorandthecorrectedAOA afterremovalof thedynamicallyinducedbiaserror. (Bottom) Correspondingtimedomainmeasurementof yawmoment.
70
71
74
75
76
77
79
80
82
83
88
89
91
93
iv
Figure5.21
Figure5.22
Figure5.23
(Top)Timedomainresponseof theAOA measuredwith theservo-accelerometersensorandthecorrectedAOA after removal of the
dynamically induced bias error. (Bottom) Corresponding time
domain measurement of yaw moment.
(Top) Time domain response of the AOA measured with the servo-accelerometer sensor and the corrected AOA after removal of the
dynamically induced bias error. (Bottom) Corresponding time
domain measurement of yaw moment.
(Top) Time domain response of the AOA measured with the servo-accelerometer sensor and the corrected AOA after removal of the
dynamically induced bias error. (Bottom) Corresponding time
domain measurement of yaw moment.
94
95
96
v
Table5.1
Table5.2
Table5.3
Table5.4
Table5.5
AppendixA Table1
AppendixA Table2
AppendixA Table3
LIST OF TABLES
Modal Parameters for Commercial Transport Model
Modal Parameters for High Speed Transport Model
Modal Parameters for High Speed Transport Model in Test Section
Summary of Wind Tunnel Results
Summary of Wind Tunnel Results for One Second Data
Acquisition Intervals
Transport model Worst Case Loading Conditions
Transport Model Natural Frequency Comparison
Transport Model Mode Radius Comparison
64
72
86
90
97
107
108
109
vi
aB(t)
a c
an(t)
at(t)
ax(t)
A fil
Ar
Aunf (t)
cij
CO
CFi
CC
CM
CM i
ccr
dcg/bc
dF/bc
di
D
LIST OF SYMBOLS
acceleration bias error estimate
centrifugal acceleration
time dependent normal acceleration
time dependent tangential acceleration
time dependent longitudinal acceleration
filtered AOA signal from inertial device
peak acceleration for r th mode
time dependent unfiltered AOA signal
damping coefficients
drag coefficient
force coefficient for degree of freedom i
lift coefficient
slope of lift coefficient versus angle of attack
pitching-moment coefficient
moment coefficient for degree of freedom i
correlation coefficient for least square fit of mode r
distance from model mass center of gravity to model balance center
distance from point of force application to model balance center
characteristic length corresponding to degree of freedom i
energy dissipation function
vii
f,
FA
Fc
FD
FL
FN
g
i
kr
m
m_j
mm
n
N
p_
Pry
qi
q_
natural frequency of r th mode in Hertz
axial force
centrifugal force
drag force
lift force
normal force
gravitational constant
scalar index
inertia about the y-axis for a reference at the model balance center
stiffness influence coefficients
bending stiffness
torsional stiffness
number of included modes
inertia coefficients
model mass
number of lumped masses used to represent the wind tunnel model system
number of degrees of freedom in the analytical model
modal coordinate for mode r
amplitude for mode ry
.thgeneralized coordinate
derivative of i th generalized coordinate with respect to time
dynamic pressure
viii
ai
an
OMi
?-
rp
ry
S
Si
T
t
U
Vr(t)
v,
V=
Yr(t)
rr
o_
O_
6
A
E
Pr
non-conservative generalized applied force (or moment) associated with qi
generalized aerodynamic force for translation degree of freedom i
generalized aerodynamic moment for translation degree of freedom i
current mode number
designates pitch plane mode
designates yaw plane mode
reference area of the model
reference area for degree of freedom i
kinetic energy of the system
time in seconds
potential energy of the system
time dependent velocity for rth mode
peak velocity for mode r
free-stream wind velocity
time dependent displacement for r th mode
peak displacement for rth mode
pitch rotation angle between undeflected sting and inertial coordinates
model attitude, or angle of attack
estimate of model attitude with bias error correction
difference
model attitude error
effective radius of r th vibration mode
ix
COl.
d
dt
_r
It]
[-,.]
[_
[_
(q}
{Q}
{Q'}
(p}
{_,}_
[_]
[_]
(x,y,z)
(xs, Ys, zs)
(x_o, Y_o, z_o)
(XBi, yBi, ZBi)
circular frequency of rth mode
derivative with respect to time
modal damping for mode r
matrix of damping coefficients
identity matrix
matrix of stiffness coefficients
matrix of inertia coefficients
vector of generalized coordinates
subset of generalized coordinates representing rigid model
vector of generalized forces
vector of generalized forces transformed to modal space
vector of modal coordinates
mass normalized modal vector for mode r
mass normalized modal matrix
diagonal matrix of natural frequency squared
modal, or eigenvector matrix
diagonalized damping matrix
inertial coordinate system
undeflected sting coordinate system
origin of sting coordinate system relative to the inertial coordinate system
body axis coordinate system for ith concentrated mass
x
(Xi, yi, Zi)
13)
position of ith body coordinate axes relative to sting axes
rotation of ith body coordinate axes relative to sting axes
AOA
BPF
c.g.
FEM
Hz
LPF
NTF
LIST OF ACRONYMS
Angle of Attack
bandpass filter
mass center of gravity
finite element model
Hertz
lowpass filter
National Transonic Facility
xi
Chapter 1
INTRODUCTION
1.1 Introduction
Model vibrations are a significant problem when testing in high pressure wind tunnels.
As discussed by Young [1], model vibrations can jeopardize model structural integrity,
overload force balances and support stings, cause models to foul, affect aerodynamic
data, and often limit test envelopes.
The National Transonic Facility [2] , NTF, is a transonic wind tunnel located at NASA
Langley Research Center which has the capability for testing models at Reynolds number
up to 140 million at Mach 1 and dynamic pressure up to 7000 pounds per square foot.
The NTF is a cryogenic facility with operating temperatures as low as -290°F. Severe
model vibrations have been encountered on a number of models since the tunnel began
operation in 1984. References 3 through 6 document studies of model and model support
vibrations in the facility. During a 1993 wind tunnel test, increased uncertainty in the
model attitude data was observed for periods of high model vibration. The response of
the onboard instrumentation to electrodynamic shaker input to the model without tunnel
airflow, "wind-off", was examined for two transport model systems [7, 8] at the NTF.
These wind-off dynamic tests found model vibration induced errors over an order of
magnitude greater than the required accuracy for the inertial model attitude
measurements.
This researchinvestigatestheeffectof wind tunnelmodelsystemdynamicsonmeasured
aerodynamicdata. The objective is to improve the aerodynamicdataquality during
conditions of high model vibrations.The equationsof motion are developedusing
Lagrange'sequationsfor the generalizedproblemof a cantileveredwind tunnel model.
This was the first time a systemdynamicanalysisapproachwas usedto examinethe
effectsof model vibrationson the aerodynamicdata. The modal solution of the
equationsof motion providesvaluableinsight into theunderlyingphysicsand provides
the basisfor the proposed"modalcorrectionmethod"for dynamicallyinducederrorsin
wind tunnel model attitudemeasurements.The proposedcorrectionmethodusesthe
modalpropertiesof themodelsystemto minimizethenumberof transducersrequiredfor
implementation. This is critical due to limited interior model spaceand thermal
considerationsassociatedwith cryogenicwind tunnelswhere heatedinstrumentation
packagesare required.The methodwas the first time domain techniquedevelopedto
compensatefor multiple modesin both the pitch andyaw planesof the modelsystem.
The ability to correct in the time domain is necessitatedby the randomnatureof the
measuredmodel dynamic responseand the increasedemphasison correlating time
dependentchangesin modelattitudewith aerodynamicloads.
1.2 Problem Description
The majority of wind tunnel tests are conducted with a model supported on the end of a
long tapered cylinder, referred to as a "sting", which is cantilevered from an arc sector or
movable vertical strut-type of support. A schematic of the NTF model support system is
2
shownin Figure 1.1. Pitchattitudeof themodelis adjustedby rotationof thearcsector.
Thearcsectorsystemis designedsuchthatthecenterof rotationof thearcsectoris at the
model,sothat changingthemodelpitchangledoesnot translatethemodelto adifferent
position relative to the wind tunnel test section. Roll attitudeof the model can be
adjustedby rotationof the sting. A six componentforcebalanceis usedasthe single
point of attachmentbetweenthe modeland supportstingas shownin Figure 1.2. In
order to achievethe desiredmeasurementaccuracyon three force and threemoment
aerodynamicloadcomponents,thebalanceis designedto be flexible ascomparedto the
sting. Theflexibility of thebalanceresultsin vibrationmodescharacterizedby themodel
vibratingasa rigid bodyona spring(forcebalance)in pitch,yawandroll. Thesemodes
are typically lightly dampedand often excitedduring wind tunnel testing [6]. Other
primarylow frequencyvibrationmodesareassociatedwith stingbendingin thepitch and
yaw planes,where most of the bendingdeformationoccursover the small diameter
portionof thestingnearthemodel.
This dissertationwill focuson the "pitch-pause"[9] wind tunneltest techniquesincethe
supportingwind tunnel testdatawere acquiredusingthis techniqueat the NTF. The
pitch-pausetechniqueis a commontestmethodusedto obtainaerodynamicloadsdatain
continuousflow, closedcircuit wind tunnels. In thepitch-pausetechnique,themodel is
movedto a prescribedangleof attackwith respectto the velocity vector,the transient
responsesareallowedto decay,andthentheforcebalance,pressure,andangleof attack
data are measured.At the NTF, the data measurementperiod is one second. The
3
Bearings
Sting
Model
Fixedfairing
Rolldrive
Locking pin -_
Arcsector
Crosshead
Shell
Insulation
Hydrauliccylinder
Figure 1.1 National Transonic Facility model support system.
Modelattach point
Balance momentcenter
X
Y
Balance/stingAOA Balance joint
package g sin(z
Figure 1.2 Schematic of wind tunnel model system.
5
procedure is then repeated for a series of model attitudes, which is referred to as a polar.
Increasing emphasis on wind tunnel productivity is pushing facilities towards shorter test
times (less time on point for transient dynamics to decay) and the effect on the
aerodynamic data accuracy must be evaluated.
During wind tunnel tests, free stream turbulence produce fluctuations in dynamic pressure
and flow angularity leading to unsteady forces on the model. The force balance and angle
of attack measurements are typically low-pass filtered and averaged to obtain "steady-
state" model attitude, aerodynamic force and moment data [10]. It is not unusual for the
peak-to-peak variation of the dynamic component of the "steady-state" force data to be
50% or more of the true mean. In Reference [11], the unsteadiness of the airflow and the
resulting model vibration is discussed. It is noted that, if the model vibration response
due to the flow unsteadiness is excessive, the ability to accurately measure the
aerodynamic quantities of interest may be compromised. Mabey [11] approaches the
problem by examining methods to reduce the flow unsteadiness in the wind tunnel. In
the Advisory Group for Aerospace Research and Development (AGARD) report entitled
"Wind Tunnel Flow Quality and Data Accuracy Requirements" [12], one of the data
accuracy issues is the measurement of, and correction for, aeroelastic deformations and
vibrations of models and support systems. Accuracy requirements [12] for lift, drag, and
pitching moment for transport type aircraft in the high speed regime are: Lift Coefficient
AC L = 0.01 ; Drag Coefficient AC D = 0.0001 ; Pitching-moment coefficient
AC M =0.001. In order to maintain the required accuracy, the tunnel free-stream
6
conditions must be repeatablewithin the following boundaries: Tunnel total and
stagnationpressure,AP= 0.1%;Model angleof attack,Ao_ = 0.01°; and Mach Number:
AM = 0.001. As an example, for conditions near a maximum lift to drag ratio, an increase
of 1 drag count (AC D = 0.0001) will decrease the payload by approximately 1% for the
long-range mission of a large transport aircraft.
The predominant instrumentation used to measure model attitude or angle of attack
(AOA) in wind tunnel testing at NASA Langley Research Center and wind tunnels
throughout the world is the servo accelerometer device described in Reference 13. The
inertial AOA package is shown installed in the nose of a test model in Figure 1.2. The
AOA package uses a servo accelerometer with its sensitive axis parallel with the
longitudinal axis of the model. For quasi-static conditions, this sensor provides a model
attitude measurement with respect to the local gravity field to an accuracy of _+0.01 ° over
a range of +20 °. An increment of 0.01 ° corresponds to an acceleration of 175 micro-g's.
During wind tunnel testing, the model mounted at the end of the sting experiences
dynamic oscillations due to unsteady flows that result in a bias error in the model attitude
measurement.
Young et. al [7] conducted an experimental study on the inertial model attitude sensor
response to a simulated dynamic environment in 1993 at the NTF. The experimental
study [7] clearly established that AOA bias error is due to centrifugal forces associated
with model vibration. For a single mode in simple harmonic motion, this is shown
7
schematicallyin Figure 1.3.TheAOA packagemovesonacirculararcabouta centerof
rotation that is mode dependent. For a single mode, the motion of the AOA package can
be treated similar to that of a simple pendulum. The centrifugal acceleration will act
outward from the center of rotation and be equal to the tangential velocity squared
divided by the radius arm. During wind-off dynamic tests, centrifugal acceleration due to
model vibration created a bias error over an order of magnitude greater than the desired
device accuracy of 0.01 degree. The bias error was found to be dependent on the
vibration mode and amplitude. The study revealed the complexity of the problem when
multiple vibration modes were present involving both pitch and yaw motions.
Although the Reference 7 study was conducted at the NTF, the AOA measurement error
due to model dynamics is not unique to this wind tunnel or to cryogenic wind tunnels.
The problem exists anytime model attitude is being measured by an inertial device in the
presence of significant model system vibrations. The amount of error in the inertial
model attitude measurement is dependent on the model system dynamics (i.e. will vary
for each model system) and is very difficult to quantify during actual wind tunnel tests.
Space limitations in wind tunnel models require that the number of additional transducers
used to implement a correction be minimized. This is illustrated by the wind tunnel
model instrumentation cavity shown in Figure 1.4. Also, in a cryogenic facility, such as
the NTF, special AOA sensor packages [13] are required. The instrumentation must be
placed in a heated package to maintain the sensors and obtain accurate and calibrated
AOApackage
a c
PF
(o r
oI. _.,
_d
X
Mode centerof rotation
Y
I___/
II
II
g a,I
II
III
I
X
Figure 1.3 Effect of vibration on inertial model attitude measurement.
C_
O
c_
o_ o
LT_
measurements at extreme temperatures (-290°F). Past experience with accelerometers
placed outside of the heated instrumentation package has revealed problems ranging from
sensitivity shifts due to temperatures variations to complete signal loss. The extreme
temperatures conditions, limited interior model space, and stringent accuracy
requirements necessitate placing the additional transducers necessary for correction of the
inertial AOA sensor output in the heated instrumentation package. The centrifugal
acceleration not only affects the inertial AOA device but can, if amplitudes are
sufficiently high, affect the desired axial force or drag measurement accuracy. The effect
of dynamics on pressure measurements can be a factor but is not addressed in this
dissertation.
1.3 Literature Review
Previous analyses of wind tunnel model system dynamics were restricted to a planar
problem. Burt and Uselton [14] examined the effects of sting vibrations on measured
dynamic stability derivatives. The equations of motion were derived for model rigid body
motion in the pitch plane using Newton's second law. Billingsley [15] uses Lagrange's
equations to derive the equations of motion for a cantilevered sting-model system. Again,
the derivation is restricted to motion in the pitch plane. Young et. al [7] have shown that
model yaw vibration can result in an error in the measured pitch angle for a model-
mounted inertial angle of attack device. Therefore, an analytical model is required that
includes both pitch and yaw plane dynamics to better evaluate the effects of model
dynamics on the measured aerodynamic data.
11
The first correctiontechniquefor modelvibrationinducederrorsin inertial wind tunnel
modelattitudemeasurementswasdevelopedin 1984by PeiterFuijkschotof theNational
AerospaceLaboratory in the Netherlands[16]. This time domain technique was
developedfor one vibration modein eachthe yaw and pitch plane. Two additional
accelerometersareusedto measurethetangentialaccelerationsdueto the yawandpitch
motion of the model. The tangentialaccelerationsare integratedto obtain velocity,
squared,and divided by a scalefactor to compensatefor the effective radius of the
vibrationmode. Thissignalis thenaddedto theunfilteredAOA outputto cancelthebias
term.Themoderadiusin theyawandpitchplaneis determinedby tuningapotentiometer
while manuallyexciting the model in the yaw and pitch plane,respectively. A major
drawbackis that this techniquedoesnot addressthecasewheremultipleyaw andpitch
modesarepresent.
Renewedinterest in the effects of model vibrationson the measuredaerodynamic
quantities was prompted by the 1993 study of Young et. al. [7]. Prior to this
investigation,only asinglemodein themodelpitchandyawplaneswasconsidered.This
studyshowedthepotentialfor multiplemodesin eachplaneto participate.Severalrecent
studieshavebeenconductedatNASA LangleyResearchCenterto examinetheeffectsof
modelvibration on model attitudemeasurementdevices[7, 8, 17, 18]. In addition,
analysisof the vibration effects on gravity sensinginclinometers is underway by
Fuijkschot[19,20]of theNationalAerospaceLaboratoryin theNetherlands.
12
Frequencydomaincorrectiontechniqueshavebeenproposedby Young et. al [7] and
Tchenget. al [18]. Thecorrectionmethodof Younget. al is derivedusingan average
displacementof the model throughone cycleof vibration. This methodrequiresthe
measurementof thenaturalfrequenciesandcorrespondingpeakaccelerationmagnitudes
from thefrequencyspectraof theyawandpitchaccelerations.Youngproposesthat the
requiredscalefactor,effectiveradius,bedeterminedempiricallyduringwind-off ground
vibration tests.The correctionmethodof Tcheng[18] requiresthe measurementof the
natural frequenciesfrom the frequencyspectraof the tangentialaccelerationsand the
secondharmoniccomponentsfrom thefrequencyspectrumof theunfilteredAOA signal.
This techniqueis difficult to implementdueto theparticipationof multiple modesand
the required data accuracyto measuresmall magnitudesat the secondharmonic
frequency.Bothtechniqueshaveimplementationproblemsdueto therequiredfrequency
domainsignalprocessingof randomwind tunnel test dataover short (1 second)data
acquisitionperiods.
Another method under developmentby Tripp [8] usestime and frequencydomain
analysesto estimateand correct for the dynamicbias error. The proposedtime and
frequencydomainbiaserror correctionalgorithmis basedon thebias term for a single
yawmodebeingrepresentedby thesquareof thevelocitydividedby themoderadius. A
sensitivecorrelationtestbetweentime seriesis providedby the crossspectraldensity
coherencefunction. Correlatedspectralcomponentscommonto boththeunfilteredAOA
signalandsquareof the dynamicyawor pitchmeasurementappearin the crossspectral
density coherencefunction. Other spectralcomponentscommonto the auto spectra
13
which arenot phasecoherent,i.e. unsynchronized,tendto be removedfrom the cross
spectrumby averagingand canceledby normalization,and do not appearin the cross
spectralcoherencefunction. Thecoherencefunctionandcrossspectrumthusprovidea
meansof detectingandquantifyingAOA biaserrorsdueto angularoscillation. Thecross
spectraldensitycoherencefunctionis examinedfor spectralcorrelationwithin the AOA
passbandand the correspondingmodal frequenciesare identified. The modal radius
correspondingto eachnaturalfrequencyis estimatedby aleastsquaresfit of the integral-
squaredyaw(or pitch) measurementto thedynamicAOA output.This requiresa longer
datarecordinitially (>_10seconds)to obtaina goodestimateof the moderadius. This
moderadiusis thenusedasaconstantfor theremainderof thedatapoints. A bandpass
filter aboutthemodalfrequencyis usedto isolateaparticularmode.Theresultingsignal
is thennumericallyintegratedandsquaredanddividedby the scalarmoderadiusto give
thebiaserrorassociatedwith a particularmode. ThecorrectAOA output is thenfound
by subtractingoff the contributionsfrom all of the modesshown to have spectral
correlationandlow-passfiltering the result. In wind-off dynamictests[8], this method
had implementationproblemsdueto significantlow frequencyrandomdisturbancesin
the integral-squaredyaw (or pitch)measurementswhich wereabsentin the AOA time
series.
After the needto compensatefor multiple vibrationmodeswasdemonstratedat NASA
Langley ResearchCenter [7,8,17], Fuijkshot extendedhis time domain correction
techniqueto compensatefor multiple modes[19, 20] usingEuclideankinematicsof a
14
solid body. This work was donein parallel with the proposedtime domain "modal
correctionmethod"that is thesubjectof this dissertation.For a givenplaneof motion,
Fuijkshotproposesmeasuringboth therotationalrateandthevelocityof therigid model
anddeterminingthecorrectiontermfrom theproductof the two signals. The rotational
rateandvelocity signalsfor theyaw(orpitch)planewill containthecontributionsfor all
modesacting in that plane. The radiusfor eachof the modeswill not need to be
determinedexplicitly. Thecorrectiontermsfor thepitchandyawplanesarethenadded
to theunfilteredAOA signalprior to filtering. Themethodis currentlyunderevaluation
andhasbeenverified for sinusoidaltests[20]. Thevelocitycanbedeterminedthrough
integration of an accelerometersignal. In application,the rotational rate has been
obtainedby integratingthe differencefrom two linear accelerometersattachedto the
model fuselage,oriented in the yaw (or pitch) plane, divided by the accelerometer
separationdistance. This assumesthe accelerometersareconnectedby a rigid model
fuselage. This correctiontechniquerequiresfour additional transducersin order to
determinethe rotational rate and velocity in the pitch and yaw planes. The limited
interior spacein modelsandextremetemperatureenvironmentsin somewind tunnels,
where heatedinstrumentationpackagesare required, may prohibit this number of
transducers.This methoddoesnot providea meansof checkingthe rigid-bodymodel
assumptionsuponwhichit isbased.
In general,theproposedtime domaincorrectionsprovideseveraladvantages.First, time
domainsignalprocessingcanbe appliedto the randomwind tunnel testdataacquired
15
over shortdatasamplingperiods. Secondly,the inertial AOA packageoutput can be
correctedfor the dynamicallyinducederrorsto give an accuratetime domain model
attitudesignal. The measurementof time varying signalsand analysisof this data is
becoming a more significant requirementfor subsonicand transonic experimental
researchers[21].Themeasurementof instantaneousandaveragevaluesof modelattitude
andcorrelationwith measuredmodelloadsis gainingincreasedinterest.
1.4 Solution Approach
The research is divided into the following four areas: examination of the effects of
models dynamics on aerodynamic data; development of a theoretical model; development
of a correction for model vibration induced errors in inertial wind tunnel model attitude
measurements; and experimental verification.
In Chapter 2, the significance of the problem is shown by examining the effects of model
dynamics on the measured drag force and corresponding drag coefficient. Errors
introduced by the centrifugal forces associated with model vibration are quantified. The
propagation of the angle of attack errors during the transformation of the measured forces
from the model body axes to the wind axes is also examined.
In Chapter 3, the governing equations of motion for a cantilevered wind tunnel model
system are derived in discrete form using Lagrange's equations. This formulation
describes both pitch and yaw plane dynamics. The equations of motion are solved using a
16
modalanalysisapproachto obtainthe generalized,modal,solution.Basedon observed
behaviorof wind tunnelmodelsystems,theproblemis simplified.
In Chapter4, thetheoreticalmodelis usedto developa time domaincorrectionmethod
for modelvibrationinducederrorsin inertialwind tunnelmodelattitudemeasurements.
The implementationof the proposed"modal correctionmethod" using digital signal
processingtechniquesis alsodescribed.
In Chapter5, themodalcorrectionmethodis verifiedthroughacombinationof wind-off
dynamictestson two transportmodel systemsand wind tunnel testdata. The modal
correctionmethodis appliedto wind-off modeldynamicresponsedata for sinusoidal,
modulatedsinusoidalandrandomshakerinputsin thepitchandyawplane. In addition,
the modal correctionmethodis appliedto measureddynamicresponsedata recorded
duringwind tunneltestingof atransportmodelin theNTF.
In Chapter6, theresearchresultsaresummarizedandrecommendationsfor futurework
aredescribed.
17
Chapter 2
EFFECTS OF MODEL DYNAMICS ON AERODYNAMIC DATA
2.1 Introduction
For wind tunnel data acquisition, Steinle and Stanewsky [12] recommend that samples of
data be taken over a time interval sufficient to average out the effects of dynamic
response and unsteady flow to establish the desired confidence interval. However, as
discussed by Buehrle and Young [17], the centrifugal acceleration created by model
vibration results in a bias error in the inertial wind tunnel model attitude measurement.
For wind-off sinusoidal model response, it is shown that the inertial angle of attack
measurement has a mean offset which cannot be removed by filtering or averaging.
Errors over an order of magnitude greater than the required device accuracy of 0.01 ° are
possible [7, 8].
In this chapter, the effect of model vibration on the force balance measurements is
quantified. The direct effect of the vibration induced centrifugal force on the accuracy of
the measured forces is examined. Typically, the forces and moments are measured by an
internally mounted strain-gage balance which has a coordinate system that is fixed to the
model. This data is transformed to obtain the desired lift and drag force components
using the measured model attitude. The propagation of the model attitude error during
the transformation process is also examined.
18
2.2 Force Balance Measurements
Model vibration induced centrifugal forces result in errors being introduced into the
balance forces. The centrifugal force F can be writtenC
Fc = mma c (2.1)
where m m is the model mass and a is the total centrifugal acceleration. It is anticipatedC
that the centrifugal force acting on the model will be small. For the servo accelerometer,
a centrifugal acceleration of .00175 g's corresponds to a model attitude error of 0.1
degrees, which is 10 times the required device accuracy. This same centrifugal
acceleration will result in only a 0.26 pound centrifugal force for a model weighing 150
pounds. For the high dynamic pressure wind tunnel tests that produce significant model
dynamics, this would result in a drag coefficient error less than the required accuracy.
2.3 Transformation of Balance Forces
A more significant error in the measured forces may occur due to errors in the measured
model attitude. The propagation of the model attitude error into the measured drag force
was described by Owen et. al. [22]. The strain gage balance forces are measured in the
model body axes, which are fixed to the model, and transformed to the lift and drag force
components using the measured model attitude. Figure 2.1 shows the relevant forces and
coordinate axes. The axial force, FA, and normal force, FN, are the balance forces
measured relative to the body axes, (XB, ZB). The lift force, FL, and drag force, FD, are
defined relative to the wind axes, (x ,z), which have one axis parallel to the flow
direction. The measured model attitude, _, defines the transformation between the two
19
F'ltFN
direction
ZB ¢ _rz L_ FA
Figure 2.1 Aerodynamic forces and model coordinate axes.
20
coordinate systems. The lift and drag forces can be written
F L = F N cos(Or ) - F A sin(or )
FD = F a cos(ct ) + F N sin(or )
If the model attitude has an error, E, the lift and drag forces can be written
FL = F N cos(o_ + _ ) - Fa sin(or + E )
Fo = F a cos(or + E ) + FN sin(oc + Iz)
(2.2a)
(2.2b)
(2.3a)
(2.3b)
The errors in the lift and drag forces due to the model attitude error, e, are defined by the
differences of Equations 2.2 and 2.3.
AF L = FN (cos0x + E ) - cos(o_ )) - FA (sin(o_ + E ) - sin(or )) (2.4a)
AF D = FA (cos(or + _ ) - cos(o_ )) + FN (sin(or + e ) - sin(o_ )) (2.4b)
Expanding the trigonometric expressions and applying small angle assumptions for e,
AF L = -e (F N sin(or )+ FA cos(o_ ))
AF D = _ ( FN cos(or ) - Fa sin(cx ))
Substituting from Equation 2.2 for the terms in parentheses results in
AF L = -EF o
AF D = eE L
The aerodynamic forces are expressed in coefficient form as
(2.5a)
(2.5b)
(2.6a)
(2.6b)
C L - EL and C D - FD (2.7)q_S q_S
where CL is the coefficient of lift, CD is the coefficient of drag, q_ is the dynamic
pressure, and S is the reference area of the model.
21
RewritingEquation2.6 in coefficientform gives
AC L = -EC D
ZXCD= ECL
(2.8a)
(2.8b)
As discussed in Chapter 1, the accuracy requirements [12] for lift and drag measurements
for transport-type aircraft in the high speed regime are: Lift Coefficient, AC L = 0.01 ;
Drag Coefficient, AC D = 0.0001. Except for conditions near zero lift, the coefficient of
drag is significantly less than the coefficient of lift [23]. Therefore, the error in drag
coefficient will be more critical with regard to its required measurement accuracy.
Assuming the lift coefficient can be represented as a linear function of model attitude,
gives
CL = CL_ O_+ Cons tan t (2.9)
where CLa is the slope of the lift coefficient versus model attitude plot. Substituting the
results of Equation 2.9 into Equation 2.8b gives
AC D = rt180 e (CLa °t + Constan t) (2.10)
where ot and E are expressed in degrees. The slope of the lift coefficient versus model
attitude plot for several characteristic wing shapes range from 0.05 to 0.1 per degree [23].
Using the most conservative, lower, value of CLa = 0.05per degree, the error in drag
coefficient versus model attitude is plotted for several values of model attitude error in
Figure 2.2. For this plot, the constant term in Equation 2.10 is set to zero. For a non-
zero constant term the lines will be shifted, however, the basic trends will be consistent
with those shown.
22
I,g
.eu
iE0
C3
0.0025
0.002
0.0015
0.001
0.0005
0
0
.... AOA Error = 0.20
.... AOA Error = 0.15
...... AOA Error = 0.10
- - -- AOA Error = 0.05
--AOA Error = 0.01
J
,f,i
f
,I
Jf ,i"
°f J° °i¢J '°I°
j° I
• J .I
j° °_° °J I °'°
jJ .I ° ..o-"
j° _' .°f .I" o''"
J ._ .-°°
J
2 4 6 8 10 12
Angle of Attack (Degrees)
Figure 2.2 Influence of angle of attack error on drag coefficient for CLo _=0.05.
23
As can be seen from Figure 2.2, significant errors in drag coefficient can occur due to the
propagation of the errors in the model attitude measurement. Model attitude errors
equivalent to those measured in wind-off dynamic tests (E>0.1 o) [7, 8] would result in an
error in the drag coefficient that is an order of magnitude greater than the required
accuracy at high angles of attack.
24
Chapter 3
THEORETICAL FORMULATION
3.1 Introduction
In this chapter, the equations of motion for a cantilevered wind tunnel model system are
derived using Lagrange's Equations [24- 26]. The Lagrange method provides a
generalized systematic energy approach for defining the equations of motion in any
convenient coordinate system. The resulting equations of motion are formulated in terms
of the generalized, modal, coordinates. Based on observed behavior of the model system
during wind tunnel tests, the analytical model is simplified. This simplified model
provides the basis for development of the modal correction method in Chapter 4.
3.2 Dynamic Equations of Motion
A lumped mass model will be used to represent the wind tunnel model and its support
system. This work extends the planar analysis of Billingsley [15] to include both pitch
and yaw dynamics of the sting-balance-model system.
In order to represent the model system during pitch-pause wind tunnel testing, three
coordinate systems are defined in Figure 3.1. The coordinate system (x, y, z) is the
inertial coordinate system with the x-axis parallel to the wind direction. The coordinate
system (xs, ys, z_) is fixed to the undeflected sting axis and has its origin at the arc sector
center of rotation. Recall from Chapter 1, that the arc sector is the movable portion of the
model support system that provides the pitch adjustment for the model. The arc sector is
25
° ,,,,I
X
N
designed such that its center of rotation is at the model, so that changing the model pitch
angle does not translate the model to a different position relative to the wind tunnel test
section. The coordinate system (x_, Ys, z_) can be defined by the location of its origin (xs0,
ys0, Zs0) relative to the inertial coordinate system and the pitch angle (_). The body axes
(xBi, yBi, zBi) are fixed to the ith concentrated mass. The position and orientation of the
body axes for the ith concentrated mass relative to the undeflected sting axes are defined
by the translations (Xi, Yi, Zi) and rotations (7i, oq, _i).
In the pitch-pause method of wind tunnel testing, the model is pitched to a desired angle
(eq) and paused to establish "steady-state" conditions. This results in:
Xs0 = J's0 = Zs0 = _s = 0 (3.1)
where the "o" denotes the derivative with respect to time. The time varying components
representing the motion of the ith mass are the translations (xi, yi, zi) and rotations (Ti, oq,
_i) relative to the undeflected sting. The generalized coordinates describing the motion of
the cantilevered sting-balance-model system can be written:
{q} ={Xl Yl Zl T1 °_ 1 _1 ... Xn Yn Zn '_n °_n _n} T (3.2)
where n is the number of lumped masses used to represent the wind tunnel model system.
The rotation angles (7i, _, 13i) induced by inertial and aerodynamic loading are small.
Therefore, in the subsequent derivations, small angle approximations (i.e., sin(o_ )----o_;
cos(o_ ) = 1) can be used and higher order terms can be neglected.
27
3.2.1 Lagrange's Equations
For a lumped mass model, Lagrange's Equations [26] can be written as:
d ( _T "_ _T _D _U
-_--j-_-+--+--=Qil[_qi_qi _qi _qi for i=l ..... N
where:
(3.3)
D = energy dissipation function
n = number of lumped masses used to represent the wind tunnel model system
N = 6*n = number of degrees of freedom
qi = ith generalized coordinate
t_i = derivative of ith generalized coordinate with respect to time
Qi = non-conservative generalized applied force (or moment) associated with qi
T = kinetic energy of the system
U = potential energy of the system
3.2.2 Kinetic Energy
The kinetic energy of the system can be written [26]:
1 NN
T = "_i_=l _. mijqiqj (3.4)J=l
where the mij are inertia coefficients. For small oscillations about the equilibrium, the
inertia coefficients are constants and the kinetic energy is a function of {q} only. The
mass matrix is symmetric, i.e., mij=mji .
_T_=0
3qi
Since the kinetic energy is not a function of {q},
(3.5)
28
Taking the derivative of the kinetic energy with respect to the time derivative of the ith
generalized coordinate gives:
OT N_ Y_mijitj (3.6)
Oqi j=l
The time derivative of Equation 3.6 is:
.-_tt(,-_qi )= jE=lmiSl j (3.7)
3.2.3 Potential Energy
For the cantilevered wind tunnel model, the potential energy is equal to the strain energy
stored in the sting-model system. A detailed derivation of the strain energy is given by
Fung [27]. The potential energy can be written in terms of the stiffness influence
coefficients as:
NN
l i_=lj_=lkt.lqtq j (3.8)U 2_-- .. . .
where the stiffness influence coefficient, kij, is the force required at point (i) due to a unit
deflection at point (j) with all other points held fixed. The stiffness influence coefficients
are symmetric, i.e., kij = kji. Taking the derivative of the potential energy function with
respect to the generalized coordinate ( qi ) gives:
o_U N
_ _., kijq j (3.9)Oqi j=l
29
3.2.4 Energy Dissipation Function
For the case of viscous damping, a dissipation function, D,
energy function can be defined [26].
N N
where the damping coefficients, cii, are symmetric, i.e., cij =cii.
the dissipation function with respect to the time derivative
coordinate gives:
OD N- Z co0 j
_)t)i j=l
analogous to the potential
(3.10)
Taking the derivative of
of the ith generalized
(3.11)
3.2.5 Generalized Forces
The primary generalized forces are the unsteady aerodynamic loads. The aerodynamic
loads will be modeled using a quasi-steady approximation [15]. The generalized
aerodynamic forces associated with the translation degrees of freedom are modeled as:
QFi = q_,,SiCF i (3.12)
where, q. is the dynamic pressure and Si is the characteristic area. The coefficient CFi
will be assumed linear and is a function of the model attitude. Similarly, the generalized
aerodynamic moments associated with the rotational degrees of freedom are modeled as:
QM i = q_SidiCM i (3.13)
where di is a characteristic length and the coefficient CMi is assumed to be a linear
function of model attitude.
30
3.2.6 Equations of Motion
The equations of motion for the ith lumped mass can be obtained by substituting the
results from Equations 3.5, 3.7, 3.9, 3.11, 3.12 and 3.13 into Equation 3.3. In matrix
form this yields:
[ M]{/_} + [C]{q} + [K]{q} = {Q} (3.14)
where,
{q} ={Xl Yl Zl 3'1 °_ 1 Ill ... Xn Yn Zn "_n O_n _n} T
[C] is a square matrix of the damping coefficients, cij
[K] is a square matrix of the stiffness coefficients, kij
[M] is a square matrix of the inertia coefficients, mij
{a} is a vector containing the generalized forces, Qi
3.3 Modal Analysis
The modal analysis technique [24, 28] will be used to solve for the dynamic response of
the multiple degree of freedom system described by Equation 3.14 with initial conditions
{q(0)}={q0} and {q(0)}={q0}. The modal analysis technique is based on the
transformation of the coupled equations of motion represented by Equation 3.14 into an
independent set of equations using the normal modes of the system.
In the modal analysis technique, the first step is to obtain the eigenvalues and
eigenvectors associated with the mass and stiffness matrices of the system. Numerical
31
methodsfor solving theeigenvalueproblemarediscussedin References21, 23 and 26.
Another approachis to obtain the eigenvaluesand eigenvectorsthroughexperimental
modalanalysis[29,30]. Oncethenaturalfrequenciesandmodeshapesareobtained,the
solutionto theeigenvalueproblemcanbewrittenas:
[M][_]['032.]= [K][W] (3.15)
where, [W] is themodalor eigenvectormatrix
['032] is adiagonalmatrixof thenaturalfrequencies,_ ,squared
Normalizingthemodalmatrixwith respectto themassmatrixyields:
[_]T [M][_] = ['i. ] (3.16a)
[_]T [K][_] = ['032.] (3.16b)
where,[_] is themassnormalizedmodalor eigenvectormatrix,and
['I. ] is the identitymatrix
Thetransformationfrom thegeneralizedcoordinates,{q}, to themodalcoordinates,{p},
canbewritten:
N
{q(t)}=[_]{p(t)}= ]_ {d_}rPr(t) (3.17)r=l
where, {_ }r is the mass normalized modal vector for mode r. Substituting Equation
3.17 into Equation 3.14, and premultiplying by [_]T yields,
t.Ftdt.l{/,}+[,02.]{p}=t.F
32
Assuming the damping is a linear combination of the mass and stiffness matrices, the
transformation will also diagonalize the damping matrix.
[CI_]T [C][cI_] = ['2403. ] (3.19)
where the modal damping for mode r can be written:
1 {_}T[c]{ _}r (3.20)_r -- 20_r
Substituting Equation 3.19 into 3.18 results in
+ ]{p}+ ]{p}--{Q,}
where
(3.21)
{Q,}_-E_,Ir{Q} (3.22)
The N independent equations corresponding to Equation 3.21 can be written as
_0r (t) + 2 4 rO_rP(t) + O_2 p(t) = Qr (t) , r=l,2 ..... N (3.23)
This is the form of a single degree of freedom system with viscous damping. Using the
transformation equation (3.17), the initial conditions can be written
{q(0)}= [_]{p(0)} and {q(0)}= [_]{p(0)} (3.24)
Premultiplying these equations by [_]T[M] and solving for the modal initial conditions
gives
pr(0) = {_ }T[M]{q(0)} and/Sr(0) = {_ }T[M]{q(0)}, for r=l,2 .... ,N (3.25)
33
The solution to Equation 3.23 can be obtained using the Laplace transform method [24].
This results in
pr(t) = 1--_-i Qr(X )e -_rO3r(t-'_ ) sin0_dr (t -Z )dxO_d r 0
+e r rtl r O cos dsin dr/ J(3.26)
where (Ddr = O r _/(1- _ r2 ) is the damped natural frequency for mode r.
For a given set of generalized forces and initial conditions, Equations 3.22, 3.25 and 3.26
can be used to solve for the modal coordinates, {p}. The solution in terms of the
generalized coordinates, {q}, can then be found from Equation 3.17. The problem is
now in generalized form and can be used to estimate and correct for model vibration
induced centrifugal accelerations. However, the problem can be simplified as developed
in the following section.
3.4 Simplified Model
Once the natural frequencies and mode shapes have been obtained, the dynamic model of
the sting-model system can be simplified based on behavior observed during wind tunnel
testing. The primary dynamic components affecting the wind tunnel model
instrumentation are in the model pitch and yaw planes [7, 8]. Since the inertial angle of
attack device has its sensitive axis parallel to the longitudinal axis of the model, the
34
deviceis notsensitiveto roll motionsaboutthisaxis. Also, theeffectsof axialmodeson
the inertial angle of attack device can be removed through filtering. Therefore, only the
pitch and yaw plane motions will be considered in subsequent derivations.
Figures 3.2 and 3.3 show measured mode shapes of a high speed commercial transport
model in the National Transonic Facility (NTF). These mode shapes demonstrate several
important characteristics common to models tested in the NTF. The lower frequency
modes (<50 Hz) of the model system are characterized by rigid body motion of the model
on the more flexible sting-balance combination. The first two modes are associated with
sting bending motion in the pitch and yaw plane. In order to achieve the desired
measurement accuracy for the "steady-state" aerodynamic loads, the force balance is
relatively flexible as compared to the model and sting. The strain gage balance systems
used in the NTF [31] are designed with flexures that separate the loads into its planar
components with minimal interactions. This results in predominantly pitch or yaw plane
motion of the model for the lower frequency modes of the system. For a given mode, the
rigid-body model motion can be defined by a translation y or z along with a
corresponding rotation [3 or t_ (see Figures 3.2 and 3.3).
3.4.1 Two Degree of Freedom Example
A two degree of freedom example will be used to define some useful properties
associated with the planar motion of the "rigid" model. The modal characteristics of the
two degree of freedom system shown in Figure 3.4 will be examined. This is similar to an
35
0
E_E
_" •a
I
>
/E_c--
0
0
N
0
°_,,q
.8
t"l
o E
._ I
IIII
N
,/
X
_h.._r
0
0
N
._..q
J_
r_
._,.qLT._
KT
KBBalance moment
center
O_
Figure 3.4 Two degree of freedom model.
38
example for vehicle suspension given by Thompson [32]. In this example, the translation
and rotation coordinates at the balance moment center will be used in defining the model
motion. Using the Lagrange Method, the equations of motion are derived. For small
angles, the equations of motion are
I-mm'mdmcg/bc -mrn'dcglbc-_Z_+IkBIybcJ_(iJ [. 0 kOT_Z}=fdF/lbc} f(t) (3.27)
where, mm is the model mass, dcg/bc is the distance from the mass center to the balance
center; dF_ is the distance from the force to the balance center; Iy bc is the inertia about
the balance center; kB is the bending stiffness; kT is the torsional stiffness; z and o_ are the
displacement and rotation from the equilibrium position; and F(t) is the applied force.
The main interest is in the form of the mode shapes. The eigenvalue problem
corresponding to Equation 3.27 can be written
0 z mrn mrn'dcg/bc_Z l (3.28)
Based on measured weight, physical dimensions, and natural frequencies of a typical
transport model system, the following constants were determined.
mm=0.3313 pound-secondZ/inch
Iy bc= 19.51 inch-pound-second 2
dc_c= 5 inches
kB = 1308 pound/inch
kT = 277089 inch-pound
39
SubstitutingthesevaluesintoEquation3.28,andsolvingtheeigenvalueproblemyields
°_1 _9.38Hz;{_}1 1-1"513lfl - 2*rt = [0.0415J
(3.29a)
°_2 -26.7Hz" {q_}2 l 1"719 lf2 - 2*n ' = 10.2955J
(3.29b)
The mode shapes are depicted graphically in Figure 3.5. Note that for each mode there is
a node (point of zero motion) about which the rigid body model rotates. The position of
this node is defined by the ratio of the translation and rotation degrees of freedom.
Scaling the modes to unit rotation gives
{q_}l :0"0415 l 1 I =
[5.82] 0.2955{- _2 }
(3.30a)
(3.30b)
where the ith mode radius, Pi, is defined as the ratio of the translation and rotation mode
shape coefficients with the modal vector scaled to unit rotation. This yields a physical
interpretation of the mode radius as the distance from the node to the reference point on
the model with the positive direction defined by the model x-axis. For this example, the
mode radius values are Pl = 36.4 inches, and P2 = -5.82 inches. The radius by definition
can be positive or negative based on the mode shape. The effect of the sign of the radius
will be discussed in Chapter 5.
40
Mode 1
_ _ Node
.............. _ -_- x
Mode 2
P2 = -5.82 inch
Z
Figure 3.5 Mode shapes for two degree of freedom example.
41
3.4.2 Extension to Multiple Degree of Freedom System
The results of the two degree of freedom example can be used to simplify the
transformation Equation 3.17. Recognizing the planar characteristics of the model
response for the lower frequency modes, Equation 3.17 can be expanded as
{q(t)}:[dP]{p(t)}= _.{f) }ryPry(t)+ Y,{_ }rpPrp(t)+ _, {t_ }rPr(t)ry rp r_ry ,rp
(3.31)
The low frequency yaw modes denoted by ry are characterized by rigid body motion of the
model. Letting {q} be the subset of the generalized coordinates required to represent the
model fuselage, yields:
{q}ry = _ry{_f}ryPry (t)
!o
0
ry 00
(t) = _.t_ _BryPryry
ry
0
- Pry
0
0
0
1Pry (t)
(3.32)
The coordinates shown represent the x, y, z, T, or, and [3 degrees of freedom for a point on
the "rigid" model fuselage.
Similarly, for the low frequency pitch plane modes, rp , the rigid body motion of the
model is approximated by
rp
!o0rEp -PrpPrp(t)= t_a rp 0
1
0
Prp (t) (3.33)
42
For a given mode,the rotationand translationdegreesof freedomin the predominant
planeof motionarerelatedby themoderadius. Themoderadiusis definedastheratioof
thetranslationandrotationmodeshapecoefficientsin thepredominantplaneof motion
with themodalvectorscaledto unit rotation. This simplifiedform of thesolution,given
by Equations3.32and3.33,will beusedto developa correctionfor vibration induced
errorsin Chapter4.
43
Chapter 4
MODEL ATTITUDE BIAS ERROR CORRECTION
4.1 Introduction
In this chapter, the theoretical model is used to develop the proposed time domain "modal
correction method" for model vibration induced errors in inertial wind tunnel model
attitude measurements. The modal correction theory and implementation procedure are
described. The proposed modal correction method extends the early work of Fuijkschot
[16] to compensate for multiple yaw and pitch vibration modes. This was the first time
domain correction technique developed to compensate for multiple modes of vibration in
the model pitch and yaw planes. A time domain correction is required due to the short
data acquisition periods (1 second) for the random wind tunnel data. This is also
important in order to meet future testing needs [21] involving the correlation of
instantaneous changes in model attitude and force balance data. The modal correction
method also minimizes the number of additional transducers required by using measured
modal properties of the wind tunnel model system. This is especially critical for models
with limited interior space and in wind tunnels that have extreme temperature conditions
where heated instrumentation packages are required.
Prior to the modal correction technique, the model attitude corrections were based on the
assumption that the instrumentation package moved on a circular arc with no detailed
analysis of the underlying system dynamics. The theoretical and experimental modal
44
analysesperformedduring thedevelopmentof themodalcorrectiontechniqueprovided
valuableinsight into thedynamicbehaviorof cantileveredwind tunnel modelsystems.
Observationof the relevantanimatedmodeshapesrevealedthat the modelmovedas a
rigid body on the more flexible sting-balancecombination. The assumptionof rigid
body modelmotion is critical to thedevelopmentof multi-modetime domaincorrection
techniques.
4.2 Modal Correction Theory
The primary generalized forces are associated with the "quasi-steady" aerodynamic loads
acting on the model. Unsteady flow in the wind tunnel results in a broadband random
input to the model system. The input for this process is not directly known or measured.
For the metallic sting-model structure, the damping is low and the system acts as a
narrow band filter passing energy (or responding) at the natural frequencies of the model
system [33]. If the modes are well separated and lightly damped, the response motion at a
natural frequency, _, will be described by the corresponding mode shape, {_ }r' with
residual effects of other modes assumed negligible.
45
The physicsof theproblemcannow be studiedby consideringthe responseof a single
modeasdepictedin Figure4.1. UsingEquation3.32,theresponsefor asingleyawmode
in simpleharmonicmotioncanbewritten
{_)(t))=
x
Y
Z
7
0
*y0
= 0
0
ry d_
" Pry (t)=f_fJry
ry
0
- Pry
0Pry0
0
1
sin(O3ry t) (4.1)
where Pry is a scalar constant related to the amplitude of motion. The reference
coordinates on the rigid fuselage will be taken at the location of the on-board inertial
angle of attack (AOA) package. The translation and rotation of the AOA package can
then be written
Yry (t) = Yry sin(O_ry t)(4.2)
1
_ry (t)= --Pry Yry (t) (4.3)
where Yry is a constant representing the amplitude of motion. Taking the derivative with
respect to time gives
Yry (t) = Vry cos(0lry t) where Vry = Yry Olry(4.4)
1
_ry (t)-- --Pry Yry (t) (4.5)
46
Yry
Modecenter of
Pry"
Figure 4.1 Harmonic motion of model at natural frequency of t.o_y.
47
The corresponding tangential and normal acceleration components, a t and a n , are:
at(t)= Yry (t)= Ary sin(COryt); where Ary =-VryO3ry (4.6)
an(t) = _ry (t)Yry (t)- --
Yry2(t)
-Pry
(4.7)
Substituting for J'ry from Equation 4.4 gives
Vry2
an(t ) ---z---_cos 2 (0_ry t)..... (1 + cos(2C0ry t))Pry 2 Pry
(4.8)
Recall from Chapter 1 that the on-board inertial AOA package uses a servo-accelerometer
with its sensitive axis parallel to the longitudinal axis of the model. The vibration
induced normal acceleration results in the AOA package sensing a centrifugal
acceleration coincident with its sensitive axis. The AOA package output prior to filtering,
Aunf , becomes:
Aun f (t) = g sin e_ + ff ry (t) _Yry (t) -- a x (t) (4.9)
The first term on the right hand side of the equation is the gravitational acceleration due
to the true model attitude, _, relative to the local vertical. The second term is the
centrifugal acceleration (from Equation 4.7) caused by the model yaw motion. The third
term represent the accelerations, ax(t) , resulting from flow induced longitudinal model
vibrations (typically greater than 50 Hz). In this equation, the positive output for the
AOA package corresponds to a positive change in angle of attack. Using the modal
48
radiusto relatethetranslationandrotationdegreesof freedom(Equation4.5) of the rigid
model,theequationcanbewritten
Yry2(t)Aunf (t) = g sins ax(t) (4.10)
Pry
Expanding Yry and using the trigonometric relations from Equation 4.8 gives
Aun f (t) = g sino_ - Vry2(l+cos(2_ryt)) -ax(t ) (4.11)
2pry
This form of the equation shows that the centrifugal acceleration for sinusoidal model
response results in the angle of attack sensor having a constant, bias, term and a harmonic
component at twice the natural frequency. The harmonic component and the longitudinal
acceleration, ax(t), can be removed by filtering. Lowpass filtering (0.4 Hz cut-off
frequency) the AOA signal yields
Vry 2Afil = g sinO_ (4.12)
2 Dry
The filtered AOA signal, Afil, has a bias error due to model vibration that cannot be
removed by filtering or averaging. From Equation 4.12, it is evident that in order to
remove the bias error, a correction method that compensates for both the amplitude of
vibration, Vry, and the mode shape, Pry, is required.
49
Model pitch vibrationcausesa similar biaserror term,wherethe tangentialvelocity is
actingin thepitchplane. If thevibrationresponseis composedof multipleyawandpitch
modes,thetotalbiaserrorwill bealinearsummationof theerrorcontributionsfor them
modes.
m Vr2Afi t = gsinot - _ (4.13)
r=l 2 Pr
Or, in terms of the peak acceleration, from Equation 4.6,
m Ar 2
Afil = gsino_ - Y_ 2 (4.14)r=l 20) r Pr
The above discussion is based on the case of continuous sinusoidal model motion. In the
wind tunnel, the data is random in nature. This results in a time varying bias error that is
dependent on the number of modes participating and the amplitudes of motion for those
modes. In order to compensate for a time varying bias errors, a time domain correction
appears to be the most suitable.
The proposed time domain modal correction technique is based on the single mode model
given by Equation 4.10. Assuming the model system behaves linearly, the total bias error
will be a linear superposition of the individual mode effects. This can be written as
Aunf(t)= gsin0_ - _ v2(t------2-ax(t)r=l Pr
(4.15)
50
whereVr(t) is the velocity (pitch or yaw plane) at the AOA location for mode r and Pr is
the corresponding mode radius. For m modes, the bias error estimate, aB(t), can be
written
aB(t)= _ v2(t)r=l Pr
(4.16)
Adding the bias error estimate to the unfiltered AOA output yields
Aunf (t)+ _ v2(t) N v2(t)- gsin0_ - ]_ ax(t)r=l Pr r=m+l Or
(4.17)
The longitudinal accelerations, ax(t), can be removed through low pass filtering. The
experimental data in Chapter 5 will show the majority of the dynamic response in the
pitch and yaw plane will be concentrated in the first four to six modes. Therefore, the
effects of the higher frequency modes (denoted by r=m+l to N) will be assumed
negligible. An estimate of the true model attitude is given by
O_(t) = sin -1
m v2(t)LP Aunf(t)+ Y_
r=l Pr
g(4.18)
where the accelerations are measured in g's and LPF designates a low pass filter with a
cut-off frequency of 0.4 Hertz.
In the modal correction technique, natural frequencies, f.or , and mode shapes, {_ }r must
first be determined. This can be done using analytical or experimental techniques. In
most cases, a detailed analytical model is not available. Experimental modal analysis
51
techniques[29, 30] havebeenusedto determinetherequirednaturalfrequencies,(Dr ,
and mode shapes, {t_ }r' of the cantilevered model systems. Recall from Chapter 3 that
the low frequency "rigid-model" modes of interest have predominant motion in the pitch
or yaw plane due to the model-balance design. This is shown graphically in Figures 4.2
and 4.3. For a given mode, the radius is estimated by assuming the fuselage moves as a
rigid body and using a least square linear regression fit of the fuselage mode shape
coefficients to determine an effective point of rotation (node). A vibration mode's
effective radius is estimated as the distance from the mode's point of rotation to the
inertial AOA sensor location in the model fuselage.
The rigid body assumption used in the mode radius estimation appears to be satisfactory
for the low frequency (<50 Hz) modes that are being evaluated. The accuracy of the rigid
body assumption can be assessed using the correlation coefficient for the linear regression
fit of the fuselage mode shape coefficients. For a linear regression fit of a yaw plane
mode (see Figure 4.2), the line estimate, Yi, is defined by
Yi = axi + b (4.19)
The correlation coefficient [34] is defined as
]_x_ yExy
n (4.20)
CC r = ]I(Y'X2- (_'x)21]_Y2(_'Y)2n n
52
Undeformed
x 5 x4Jr ÷
Y5 Y4
AOAposition
x2
Y2
Modecenter of
rotation--_ _ YT
Mode radius
P
Yi = a*xi + b
Figure 4.2 Yaw plane mode of model system.
53
AOAposition Mode radiusP
Zl 4....--I-
x 3 x2 Xlx 5 x4
Undeformedz i =a.x i + b
Mode / xcenter of/rotation -'
Figure 4.3 Pitch plane mode of model system.
54
The correlation coefficient is always between -1 and +1. For values close to zero, there is
no linear relationship. For values near _+1, there is a very strong linear relationship. In
Chapter 5, the correlation coefficient is used to assess the linear regression fit of the
measured fuselage mode shape coefficients.
A second assumption is that the mode shapes do not change significantly under the wind
tunnel test conditions. This enables wind-off estimates of the mode effective radii to be
used for correction of the model attitude measurement during wind tunnel testing. In
Appendix A, the effect of aerodynamic forces on the measured modal radius were
evaluated using a finite element model of a cantilevered wind tunnel model system. The
aerodynamic forces were applied to generate a prestressed model and then the
eigensolution was performed for this prestressed loading condition. For the largest
aerodynamic forces measured on a representative transport model in the National
Transonic Facility, the predicted shifts in the modal radius were less than 4%, which is
negligible.
4.3 Modal Correction Implementation
Once the effective radius and natural frequency are obtained for each mode of interest, the
next step in the modal correction technique is the on-line measurement of the unfiltered
AOA signal, and the lateral and normal accelerations at the AOA location. Due to the
model attitude accuracy requirements ( _+0.01 ° over a range of +20 ° ), a 16-bit analog-to
55
digital converteris requiredfor thedataacquisitionsystem.Oncethe datais acquired,
thedigitizedmeasurementsareprocessedoff-line usingMATLAB ®[35].
A flow chartof thedataanalysisroutineis shownin Figure4.4. Thelateralandnormal
accelerationmeasurementsarenumericallyintegratedusingthetrapezoidalrule [36] and
scaledto obtainthe lateraland normalvelocity, respectively. The velocity signalsare
squaredusing array,or elementby element,multiplication. For each lateralmode of
interest,a linearphasefinite impulseresponsefilter is usedto defineapassbandaboutthe
natural frequency. This isolatesthe velocity squaredcomponentsof the individual
modes. Thefilters areappliedin boththeforwardandreversedirectionsto obtainzero-
phasedistortionanddoublethefilter order. This is critical for a time domaincorrection
wherethe phaserelationshipof the unfilteredAOA signaland the lateral and normal
dynamicresponsemustbemaintained. Thesquaredvelocitycomponentsfor eachmode
are divided by their correspondingmode radius and then combined using linear
superpositionto give theestimatedbiaserrordueto lateraldynamics.This procedureis
then repeatedfor the normal,or pitch, modesto determinethe bias error due to pitch
dynamics. The errorsdue to the lateraland pitch dynamicsare thencombinedusing
linear superpositionto yield thetotal biaserror. Thebiasestimateis thenaddedto the
unfilteredAOA and the result is filtered with a 0.4 Hz lowpassfilter asdescribedby
Equation4.18. Thisgivesacorrectedtimevaryingmodelattitudesignalthatcanbeused
to determinetheinstantaneousor meanangleof attackoverthedataacquisitionperiod.
56
Start )
[ mp _pb_b i=l,nap
read accel, arraysay, a_, A..f
Vy_-integral(ay) ]Vz_-integral(az)
Vy2_.Vy.,VyVz2_" Vz'* Vz ]
()
apply ban@ass filter toisolate mode effect
Vyi2 (-BPFfy i( vy 2 )
1!
estimate ith mode bias [
mBy i_-Vyi 2/121yi I
Figure 4.4 Flowchart of modal correction method.
57
.%(".%y+.%,p
apply bandpass filter toisolate mode effect
Vpi2 4[-BPFfpi(Vp 2 )
estimate ith mode bias
mBpi_-Vp i2/Ppi
1AcC'A.n_Aa
&:fa4"LPFo.4Hz(&:)
AOA(- asi n(Acr a)* 180/pi
Figure 4.4(continued) Flowchart of modal correction method.
58
Chapter 5
EXPERIMENTAL VERIFICATION
5.1 Introduction
In this chapter, the modal correction method is verified through a combination of wind-
off dynamic tests on two transport model systems and wind tunnel test data. The modal
correction method is applied to wind-off model dynamic response data to compensate for
model vibration induced errors in the inertial model attitude measurement for defined
shaker inputs in the pitch and yaw plane. In addition, the modal correction method is
applied to measured dynamic response data recorded during wind tunnel testing of a high
speed transport model in the National Transonic Facility (NTF).
5.2 Wind-off Dynamic Response Tests
This section will describe the test setup and results of wind-off dynamic response tests on
two transport models [7, 8]. The modal correction method is validated for sinusoidal,
modulated sinusoidal and random inputs to the model in the pitch and yaw plane.
5.2.1 Test Setup and Procedure
Wind-off dynamic response tests were conducted on two transport models [7, 8] in a
model assembly bay at the National Transonic Facility. The test setup for the high speed
transport is shown in Figure 5.1. The mounting consists of a "rigidly" supported
cantilever sting that is positioned by a pitch-roll-translation mechanism. The model is
attached to the sting through a six component strain gage balance.
59
c_
c_
0E
°_,.q
E_
,_,.qLr_
The model was instrumented with an inertial AOA package [ 13] maintained at a constant
temperature of 160°F. The signal conditioner for the AOA package provides both an
unfiltered, "dynamic", 0 to 300 Hz bandwidth signal, and filtered, "static", 0 to 0.4 Hz
bandwidth signal. Two miniature accelerometers were installed on the face of the AOA
package to measure yaw and pitch motions. In addition, accelerometers were installed at
several locations on the model fuselage and sting to measure the dynamic response and
natural mode characteristics.
An experimental modal analysis was performed on the model systems. Frequency
response function data were acquired for point force excitation and transferred to a
personal computer. The STAR e [36] modal analysis software was used to determine the
modal parameters from the measured frequency response functions. A least square fit of
the fuselage mode shape coefficients was used to estimate the mode radius and
corresponding correlation coefficient (see Chapter 4).
For the dynamic response tests, an electrodynamic shaker was used to excite the model
system through a single point force linkage as shown in Figure 5.2. Due to the desired
high vibration amplitudes, the model surface was protected with tape and safety wire was
used in case the glue attaching the force mounting block failed during testing. The
excitation was applied in the pitch and yaw planes at the model fuselage hard points.
Sine, modulated sine and band limited random shaker input were used. A Hewlett
Packard (HP) 3566A dynamic signal analyzer was used to provide the shaker stimulus
and record the shaker force input, model force balance outputs, AOA static and dynamic
61
_iii!iiiii!iiii
¢)
r_O
¢)
e_
.,,._
outputs, and model accelerations. This system was used to monitor the model yaw and
pitch moments which established the dynamic test conditions for acquiring model attitude
measurements. Data was also recorded using a 16-bit Analog to Digital Converter (ADC)
board in a personal computer.
The model was set at a prescribed angle of attack under static conditions. The model
system natural frequencies were identified using sine sweep excitation in the pitch and
yaw planes. For each natural frequency of interest, a sinusoidal forced response test was
conducted by controlling the shaker input amplitude to provide a defined peak to peak
pitch or yaw moment on the model force balance. The control test variables were pitch
moment for modes that had predominantly pitch motion, and yaw moment for modes that
had predominantly yaw motion. The model attitude was measured at a series of moment
amplitude levels for sinusoidal excitation at a prescribed natural frequency of the model
system.
In addition to the sinusoidal forced response tests, the high speed transport model
dynamic response was measured for modulated sine and random excitation. The
modulated sine and random excitations and responses are more representative of the
model dynamics observed in actual wind tunnel tests. The majority of the modulated
sine tests were conducted with a 0.25 Hz modulation of the first natural frequency in the
pitch and yaw planes. In each case, the inertial AOA package was used to measure the
model attitude for a series of moment amplitude levels.
63
5.2.2 Commercial Transport Model Test Results
During wind tunnel tests, the commercial transport model had significant yaw vibrations
at 14 Hertz. Discrepancies in the aerodynamic data provided the stimulus for the
investigation of the AOA device [7] and its sensitivity to model vibrations. The AOA
investigation concentrated on the first four modes. An experimental modal analysis was
conducted on the model system and the results are tabulated in Table 5.1. Figures 5.3 and
5.4 show characteristic yaw plane modes described by sting bending and balance rotation.
The mode radii and corresponding correlation coefficients are also listed in the table.
Recall from Chapter 4 that correlation coefficients near +1 indicate a very strong linear
relationship. The correlation coefficient for the least square fit of the fuselage mode shape
coefficients shows the appropriateness of the linear regression fit and validates the rigid
body model assumption for the tabulated modes.
Table 5.1
Modal Parameters for Commercial Transport Model
Mode Frequency Damping Radius Corr.
No. (Hz) (%) (Inch) Coeff.
1 10.3 1.01 38.2 .9998
2 11.2 1.78 70.5 .9971
3
4
14.4
16.5
0.46
0.59
7.05
12.0
-.9973
.9998
Mode Description
Sting Bending-Yaw Plane
Sting Bending-Pitch Plane
Model Yaw on Balance
Model Pitch on Balance
64
"0
_ E
_r r
0
0°l,-q
t_
c5
° ,,..q
_0
e_
_r_
_0°_,.q
LT_
\r
0
0.,-_
I-i
°,,=wp,
d
,.Q
0
b_
0
N
.,=,_
LI.
For the AOA investigation, the model system was locked at near zero degree angle of
attack under static conditions. Single frequency forced response tests were conducted by
controlling the shaker input to provide a defined peak to peak pitch or yaw moment on
the model force balance. The test variable was yaw moment for modes that had
predominantly yaw motion, and pitch moment for modes that had predominantly pitch
motion. The AOA response data and model accelerations were recorded for several
moment levels. This data was transferred to the MATLAB ® [35] program for application
of the modal correction technique. The measured mean AOA output, estimated bias, and
corrected mean AOA output, after application of the modal correction technique, are
shown versus balance moment in Figures 5.5 through 5.8. Recall that for sinusoidal
input, the model vibration creates a bias error or offset in the mean value. After
application of the modal correction technique, the error is reduced to the AOA device
accuracy of +0.01 degrees for all measurements except the second pitch mode. For this
case, an order of magnitude reduction is obtained.
The accuracy of the correction for the pitch axis tests may be improved by locating the
accelerometers adjacent to or inside the heated AOA package. The pitch plane
accelerometer on the face of the AOA package failed early in the test. A triax set of
accelerometers located externally on the fuselage upper surface was subsequently used to
obtain the off-axis accelerations required for the modal correction technique.
67
0.02
0
"_m -0.02
g-0.04
i -0.06
-0.08
-0.1
400
_""" _- ... ......... • .............. &....-" .... .--''_
_...__..-__._..._:-_-...............................
-- • -- Estimated Bias "X_\
- - _r -- Corrected "X_\
° \I I I
800 1200 1600 2000
Yaw Moment (Inch-Pounds)
Figure 5.5 Measured mean AOA, estimated bias, and corrected mean AOA versus
yaw moment for sinusoidal input at 10.3 Hz.
68
A
D1Oo
u
,¢
O
e
t,-,¢
0.02
-0.02
-0.04
-0.06
-0.12
-0.14
._.. . . ...... dr .......... • ........... • ............................................
I'" _''" Corrected "_\
I .... N°minal +'01° "_
1.... Nominal-.01 ° _
-0.16 , ' '
1000 4000 6000 8000 10000 12000
Yaw Moment (Inch-Pounds)
Figure 5.6 Measured mean AOA, estimated bias, and corrected mean AOA versus
yaw moment for sinusoidal input at 14.4 Hz.
69
0.02
-0.04
400
m
I-...... ...-A ........... • ......,-... ,.: :, - - _- .......... • .....
"B.......... .'_,.=,,. ............................. _.._
----e--- Measured
- •- Estimated Bias
- - _" • Corrected
.... Nominal +.01 o
.... Nominal-.01 °J I I I
800 1200 1600 2000 2400
Pitch Moment (Inch-Pounds)
Figure 5.7 Measured mean AOA, estimated bias, and corrected mean AOA versus
pitch moment for sinusoidal input at 11.2 Hz.
70
0.05
-0.05
"_ -0.1
-o.15
i -0.2-0.25
-0.3
-0.35
-0.4
'-"--';:-'- ........-'- "-"• ,r-'-'---.:---'.'.-.'--::-::-_.'.--;-.-"o
Measured _" ,,
- •- Estimated Bias -,,'_"11
- - _- - "Corrected "_ \
I I I I
4000 6000 8000 10000 12000 14000
Pitch Moment (Inch-Pounds)
Figure 5.8 Measured mean AOA, estimated bias, and corrected mean AOA versus
pitch moment for sinusoidal input at 16.2 Hz.
71
5.2.3 High Speed Transport Model Test Results
A high speed transport model system that experienced high levels of vibration [6] during
previous wind tunnel tests was selected to further investigate the effects of dynamics on
the inertial AOA package. Measurements taken during wind tunnel tests indicated that
the primary modes being excited were at approximately 8-10 Hz and 28-30 Hz [6]. An
experimental modal analysis of the model system was conducted and the results are listed
in Table 5.2. The radii and corresponding correlation coefficients for the vibration modes
were estimated using a least square linear regression fit of the modal deformations as
described in Chapter 4. The correlation coefficient for the least square fit of the fuselage
mode shape coefficients shows the appropriateness of the linear regression fit and
validates the rigid body model assumption for the tabulated modes.
Table 5.2
Modal Parameters of High Speed Transport Model
Mode
No.
Frequency
(Hz)
Damping
(%)
Radius
(Inch)
6
9.0
9.2
20.5
21.7
29.8
34.9
1.32
1.68
2.75
2.70
2.28
2.59
31.0
30.2
0.18
-1.07
-7.16
-7.65
Corr.
Coeff.
.9997
.9997
-.9993
.9999
-.9983
0.9999
Mode Description
Sting Bending-Yaw Plane
Sting Bending-Pitch Plane
Model Yaw on Balance
Model Pitch on Balance
Model Yaw on Balance
with Sting Second Bending
Model Pitch on Balance
with Sting Second Bending
72
It is important to note that the mode radius may be positive or negative dependent on the
vibration mode shape. Previously, this bias error was described as a "sting whip" [13]
error and associated with the first sting bending modes in the pitch and yaw planes. The
analyses and experimental data presented in this dissertation show that the model system
dynamics is more complex than previously assumed. The physical interpretation of the
sign of the radius is more easily understood by examining the 9.0 Hz and 29.8 Hz yaw
modes shown in Figures 5.9 and 5.10. For the case where the radius is negative, the point
of rotation for the vibration mode is forward of the AOA package. A positive radius is
defined for a point of rotation aft of the AOA package.
The significance of the sign of the radii is that the bias error may be positive or negative
dependent upon the vibration mode being excited. This is demonstrated by the response
of the two yaw plane modes shown in Figures 5.11 and 5.12. For the 9.0 Hz yaw mode,
the indicated model angle change is negative when the model is being driven with
sinusoidal excitation at the natural frequency and then returns to its nominal angle when
the shaker system is shutoff. The 29.8 Hz yaw mode, which has a negative radius value,
shows an indicated positive angle change when the model is being driven with sinusoidal
excitation at the natural frequency and then returns to its nominal angle when the shaker
system is shutoff. The excitation system was adequate to show the above trends,
however, only the first mode in each the yaw and pitch planes were excited to levels that
showed significant shifts in the indicated model attitude from the onboard inertial AOA
package. Difficulty in driving the higher frequency modes is attributed to the rigid
73
r
0
0. i,-.q
L.
. ....q
N=o.
.8
° ,,....q
r_
(D
E_o E
/l:nr-
,D
09
I
I
I
I
>,
0
o
,.o°,.,_
N
t",l
0
0
0
Inertial AOA
4.34 .............................................. ............................................ .......................................................................................................................................
Deg I
4.14i i : .......... !
0 Sec 8 Sec
Yaw Acceleration
1.5 .......................... !_........................ ............ ......
g
-1.5.....................................i..................................................................................i.....................................i.............................................0 Sec 8 Sec
Yaw Moment
3600 ............................................. ............................................ ...................... ...............................................................................................................
In-Lbs
-2400 _: ................. i.............................0 Sec 8 Sec
Figure 5.11 Inertial AOA measurement, yaw acceleration, and yaw moment versus
time for 9.0 Hz sinusoidal input in yaw plane.
76
0.4g
-0.40 Sec 8 Sec
Yaw Moment
2400 ....................i...........................................:................................................................................................................................................ii
In-Lbs
............ ].......................
-600
0 Sec 8 Sec
Figure 5.12 Inertial AOA measurement, yaw acceleration, and yaw moment versus
time for 29.8 Hz sinusoidal input in yaw plane.
77
backstop support in the model assembly bay. During previous wind tunnel tests [6], the
model coupled with the model support structure resulting in high dynamic yaw moments
with energy in the 28-30 Hz band. This points out the need to do dynamic testing with
the model installed in the tunnel.
The results of sinusoidal excitation tests for the first mode in each the yaw and pitch
plane are shown in Figures 5.13 and 5.14. The model was set at a nominal angle of 0 ° for
these tests. For a set excitation level, time domain data were acquired and stored using
the dynamic signal analyzer. These data were transferred to a personal computer where
the modal correction technique, implemented in an m-file in the MATLAB ® [35]
language, was used to estimate the bias error in the inertial device. This procedure was
repeated for several excitation levels as defined by the moment amplitude level.
As shown in Figures 5.13 and 5.14, the estimated bias error is in good agreement with the
indicated mean angle change measured with the onboard inertial AOA sensor. After
application of the modal correction method, the bias error is reduced from a maximum of
-0.146 ° to -0.009 ° for the first mode in the yaw plane and from -0.175 ° to-0.006 °
for the first mode in the pitch plane. These corrected mean angle of attack values are
within the AOA accuracy requirement of 0.01 °. Similar results were obtained for
sinusoidal input tests with the model set to nominal angles of 4.3 ° and 6 °.
78
0.02
A
t_
0
o
0
_.eo}e-
-0.02
-0.04
-0.06
-0.08
-0.1
-0.12
-0.14
-0.16
9OO
__S._ _-_*.__;_L; -::-._.--._:: _:: ._':----;--.-...
- • - Estimated Bias "X_\
oreC;%o.... Nominal -.01 o
I i I
1800 2700 3600 4500
Yew Moment (Inch-Pounds)
Figure 5.13 Measured mean AOA, estimated bias, and corrected mean AOA versus
yaw moment for sinusoidal input at 9.0 Hz.
79
0.02
A
w
2o
um
0
e
i-<
-0.02
-0.04
-0.06
-0.08
-0.1
-0.12
-0.18
1000
m
.......... ,- -- : ;:.,:._.:: ._.::_,_-,_ .........
Measu re_ _ \ \
-O--_--M_amSU:: Bias "_
- - _- - • Corrected
I I, I
2000 3000 4000
Pitch Moment (Inch-Pounds)
50O0
Figure 5.14 Measured mean AOA, estimated bias, and corrected mean AOA versus
pitch moment for sinusoidal input at 9.2 Hz.
80
In order to obtain a corrected time domain angle of attack measurement that can be used
for instantaneous or average values, it is important to maintain the phase relationship
between the measured and estimated bias error. To verify that the modal correction
method maintains this phase relationship, the bias error was examined for modulated
sine and random inputs. The measured response for modulated sine and random inputs is
also more representative of actual wind tunnel test data.
Figure 5.15 shows the measured angle of attack and estimated bias error as a function of
time for a 9.2 Hz pitch excitation with a 0.25 Hz modulation. Excellent agreement is
obtained with the difference between the measured angle of attack and estimated bias
error being less than 0.005 ° . Modulated sine tests were conducted at several excitation
amplitude levels for the first mode in each the y and z axes and consistent results were
obtained between the measured angle of attack and predicted bias errors for all cases.
In addition, the response of the AOA package for two levels of random excitation in the
pitch plane were also examined. Figure 5.16 shows an eight second record of the inertial
AOA sensor response for the highest level random excitation. The random response
measured by the pitch accelerometer on the face of the AOA package was composed of
primarily 9.2 Hz response. The bias error estimate based on only the 9.2 Hz mode
contribution is also shown in Figure 5.16. Again, the measured angle of attack and
estimated bias error are in very good agreement.
81
"O13_
I
/I'
cO
E)
I I(3O
(/)13
F:
O4
i 00 _ 04
d d
(seeJ6ec]) VOV
&m
(D
(/)
c00
•_'- if)
i i 00 0 0 00 0 00 0 0
! I
E.m
(sq7-ul) luewo_ 4O1!d
,4O
..=
OENp:V7C4d
J=
C_
O
(J
°_
O
r4 E_
°,--4
I= c__J
&)
0 =< o
_ 0
_, 0
0 0
t_
£',1
/!
O
CO
_5
I,
Oo
d!
04
(seeJ6eG) VOV
O
V
E.m
E
w
(23
_O
(/)"0c-O0
I 00 0 00 00 0O4 O4
!
E.m
!--
(sql-Ul) luewoR 401!d
O
O
© =<&
O
,-.oO O
_4
°,_
oo
5.3 High Speed Transport Model Wind Tunnel Tests
Dynamic response studies were conducted on a high speed transport model installed in
the test section of the NTF. The dynamic response characteristics were also recorded for
high speed (Mach=0.95) wind tunnel runs.
5.3.1 Test Setup in Wind Tunnel
The model was instrumented with a re-designed inertial AOA package that has two servo-
accelerometers for measuring model AOA and two dynamic accelerometers to measure
the accelerations tangent to the sensitive axis of the AOA sensors. The package is
maintained at a constant temperature of 160°F. The signal conditioner for the AOA
sensors provide both an unfiltered, dynamic, 0 to 300 Hz bandwidth signal and a filtered,
static, 0 to 0.4 Hz bandwidth signal.
Initial wind-off dynamic response studies were performed in the wind tunnel test section
using shaker excitation of the model with the arc sector in a fixed position. For the wind-
off shaker excitation tests, six additional accelerometers were mounted external to the
model fuselage to measure model yaw and pitch motion at three locations.
Data were acquired using a 16 channel digital data acquisition system with 16-bit
resolution. All dynamic signals were filtered to 100 Hz prior to recording. Data were
recorded at 200 samples per second per channel. Recorded channels included the dynamic
and static inertial AOA outputs, the tangential accelerations in yaw and pitch, and the six
84
forcebalancecomponents.Datawererecordedfor both the wind-off shakerexcitation
testsandthehighspeedwind tunnelruns.
Forthewind-off shakerexcitationtests,a HewlettPackardmodel3566Adynamicsignal
analyzerwasusedto providetheshakerstimulusandperformon-linetime andfrequency
domainsignalanalysis.The 16channelsignalanalyzerwasusedto monitorand record
theshakerforceinput,andtheresponseof thesix accelerometersmountedexternalto the
modelfuselage.
The shakerexcitationtestswereperformedwith themodel installedin the test section
andthearc sectorin a fixed position.An electrodynamicshakerwasusedto excitethe
model in theyaw planethrougha singlepoint force linkage13 inchesaft of the model
nose. Due to scheduleconstraints,theforcedresponsetestswereconductedin the yaw
plane only. The model systemnatural frequencieswere identified using sine sweep
excitation. ThedynamicandstaticinertialAOA outputs,thetangentialaccelerationsin
yawandpitch,andthesix forcebalancecomponentswererecordedfor a seriesof shaker
forceamplitudelevels for sinusoidalexcitationat a prescribednaturalfrequencyof the
model system. In addition to the sinusoidalforced responsetests, modulatedsine
excitationtestswereperformedfor a seriesof shakerforce levels. The modulatedsine
excitationsand responsesaremore representativeof the model dynamicsobservedin
actualwind tunneltests.
85
For agiventestcondition,timedomaindatawereacquiredandstoredon the 16-channel
dataacquisitionsystem.Thesedataweretransferredto apersonalcomputerwherea
softwareroutineimplementingthemodalcorrectionmethod,writtenasanM-file in the
MATLAB ®[35] language,wasusedto estimateandcorrectfor thebiaserror in the
inertial device.
5.3.2 Dynamic Response Tests in Wind Tunnel
An experimental modal analysis was performed for a high speed research model installed
in the NTF wind tunnel and the dominant modes are listed in Table 5.3. The model was
configured differently than in previous wind-off vibration tests, therefore, the modal
characteristics are different than those presented in the previous section. The mode radii
and corresponding correlation coefficients are also listed in the table. The correlation
coefficients again confirms the rigid body model assumption.
Table 5.3
Modal Parameters for Survey of High Speed Transport Model in Test Section
Mode
No.
1
2
3
4
5
6
Frequency
(Hz)
7.3
9.8
12.1
16.9
17.2
21.1
Damping
(%)
0.46
0.28
0.51
1.3
1.0
0.36
Radius
(Inch)
37.8
31.8
8.71
-0.93
-3.40
-9.54
Corr.
Coeff.
.9992
.9995
.9995
-.9985
-.9998
-.9994
Mode Description
Sting Bending-Yaw Plane
Sting Bending-Pitch Plane
Sting/Model Yaw
Model Pitch on Balance
Model Yaw on Balance
Model Yaw, 2nd Sting
Bending
86
Theresultsof sinusoidalexcitationtestsfor thefirst mode(7.3Hz) in theyawplaneare
shownin Figure5.17. This figure showstheangleof attackmeasuredwith theprimary
servo-accelerometersensorand the correctedangle of attack after removal of the
dynamicallyinducedbiaserror. Thesetestswereconductedwith themodelat a nominal
angleof 6.01° and the arc sectorin a fixed position. After applicationof the modal
correctionmethod,the error is reducedfrom a maximumof -0.087° to +0.003° for the
first modein theyawplane. As shownin Figure5.17,thecorrectedAOA measurements
are within the AOA accuracy requirement of +/- 0.01 o. The higher frequency modes were
not excited to high enough levels to produce significant shifts in the AOA measurements
during the wind-off vibration tests.
In addition to the sinusoidal tests, the bias error was examined for modulated sine input.
Figure 5.18 shows the angle of attack measured with the primary servo-accelerometer
sensor and the corrected angle of attack after removal of the dynamically induced bias
error. This data was obtained for excitation at the 7.3 Hz natural frequency with a 0.5 Hz
modulation. The corresponding measured yaw moment is also shown in Figure 5.18 and
has a maximum peak-to-peak value of 2400 in-lbs. Excellent correction is obtained
using the modal correction method with errors as large as -0.091° being reduced to less
than +/- 0.005 ° from the nominal angle. Modulated sine tests were conducted for the first
mode at several excitation amplitude levels and consistent results were obtained. For this
type of model response, correction for the dynamically induced errors results in a shift in
the mean value and a reduction in the variance of the signal.
87
6.02
,01 .............................................. " ......
6
_ 5.99
5.98
5.97
o 5.960
< 5.95
5.94
5.93
5.92
0 400 800 1200 1600 2000 2400 2800
Yaw Moment (Inch-Pounds)
Figure 5. I7. Measured and corrected angle-of-attack for sinusoidal excitation at 7.3Hz.
88
--f_'-r 1 CO , CO
_0"OcOo
V
(DE
.m
I--
O4
m
m
i
L
w
im
w
i
E
L
(.(3
cOo
O3
O4
' OO O OO OO OO4 O4
I
E.m
I--
(sq7-ul) _uewo_ Me,&
0.,=
0
N
_5
_q=:o%
0
(J
O ._
e".,-,I
O _
_o
_ O
O 0
o6
5.3.3 Wind Tunnel Test Results
The data for the first 64 seconds of a test on the high speed transport model (Mach=0.95,
Q=1800 pounds-per-square-foot, T=-254°F) were used to evaluate the dynamic response
characteristics of the primary AOA sensor and the proposed modal correction method.
The filtered output of the primary AOA is shown in Figure 5.19. Data analysis was
restricted to the periods where the model pitch angle was paused to obtain "steady-state"
aerodynamic data. The pitch acceleration was significantly lower than the yaw
acceleration over the data analysis period. Analysis of the model yaw acceleration
showed primarily 7.3 Hz response with additional energy at the 12.1 Hz natural
frequency. Intermittent response at other frequencies was observed. Initial application of
the modal correction method included the modes in Table 5.3. The AOA mean value and
standard deviation over each pause period are listed in Table 5.4.
Time Period
(Seconds)
Table 5.4
Summary of Wind Tunnel Results
Measured
AOA Mean
(Degrees)
Measured AOA
Standard
Deviation
(Degrees)
Corrected
AOA Mean
(Degrees)
-3.5403
Corrected AOA
Standard
Deviation
(Degrees)
0.01210to 9.25 -3.5664 0.0179
12.5 to 32.5 -2.5094 0.0203 -2.4764 0.0082
40.75 to 52.5 -1.4803 0.0248 -1.4392 0.0088
56to 64 -0.9308 0.0094 -0.9121 0.0057
90
tOd
I
I I I I I 1
'N""
I
,,I I
I I I
(seeJ6eO) VOV
0
0
0tO
A
0(D
03
.g
0O4
0
0
!
0
1=0
aO
0
"ID
,,-q
0
0
#
<0.<
U
<
?
<
i:m°_,_
Figures 5.20 through 5.23 show the time domain response of the angle of attack measured
with the primary servo-accelerometer sensor and the corrected angle of attack after
removal of the dynamically induced bias error for each pause period. There were no
optical measurements to confirm the corrected AOA measurements.
Since the response was primarily at the 7.3 Hz and 12.1 Hz natural frequencies, which
have positive radii, trends consistent with the wind-off modulated sine test are expected.
For a mode with a positive radius and fluctuating amplitude of motion, correction for
dynamically induced errors will result in a positive shift in the mean value and a reduced
variance for the corrected signal. The significant reduction in the variation observed in
the corrected time domain AOA signal (Figures 5.20-5.23) as compared to the measured
primary AOA and the corresponding reduction in the standard deviation for the corrected
AOA measurement indicate successful application of the modal correction method. The
periods from 12.5 to 32.5 seconds and 40.75 to 52.5 seconds (part of which are shown in
Figures 5.21 and 5.22) are the best indicators of the amount of bias reduction possible.
The inclusion of more natural frequencies in the modal correction method may aid in
improving the bias correction. It is also important to note that the low frequency
fluctuations in the corrected AOA signal may be due in part to oscillatory changes in the
model pitch attitude.
92
/
/
tO
_5!
LO
U)"0c0o
O4
I................-- T"
LO CD
!
E°m
(seeJ6eCl) VOV
U3
u)
c-OO(D
CO00
CD
Cq
E.m
I T =-
O O OO OO Ou3 tO
I
(sq7-ul) ;UeLUO_ Me),
oE
© o<E
o o
£'a
o
I
O .,-q
O
_=£[J
<mO'-"<oP=o_
O
_7,._ >_,
"t::l "
[--_
O[._-
O
c5C4
U7
C_
_T r
0
if)
oo
kz
r,-T--"
CO
I.O LO I.O CO I-
' e_i 'I I
(seeJ6eCI) VOV
O4O4
O4
OO4
or)
oo
kz
CO
i LOO O O _-O OO OLO LO
I
(sqq-ul) luem0pl MeN
0
© o
0 0
o'a
O
'_ 0
_°<mo'-"
0 _
0 _
_ ""'_
o._, ...._ _
[..,_
0 _[..u,_, 0
,,.,.,¢
!
1
/
f
\
k
/
)
)f
\
!
\
J
J
/I
.J/ __\ u_
I I I
LIIi
!
!
(seeJ6eG) VOV
O4LO
0LO
oo
O4
04LO
0
00
C/3
i O40 0 0 "_f"0 00 0LO
!
(sq'l-Ul) luemo_l MeN
0
© o<E
£'a
, _
sg
"_ 0
_ 0
0"-"
0
0 _
_ • "'_
,I::1 •
0 _[.-, u..
Nt",l
\
!
"x
#,
O4
c_13c0o
0
E
U3
_0
d '!
(seeJ6e(]) VOV
O4
"t_c0o
0
Ei--
0 0 0 0 u30 0 00 0 0u3 u3 0
!
(sql-Ul) _,uewo_ MeA
oE&.q
0 o<_
0 0
£'_
_ 0
0 • ,,,,_
o_,,I
"_ 0_r..)
<mo'-"<_
°,,-I
o,,_ _
[-.,_,---, 0
t",l
For this test, the data acquisition periods were longer than normal. The steady state data
at NTF is typically taken over a 1 second period. Differences between the corrected and
measured mean values over a given one second interval may be much larger than those
shown in Table 5.4. The results for one second intervals from 16 to 22 seconds are listed
in Table 5.5. Differences between the measured and corrected AOA mean value as large
as -.064 ° are observed over the selected one second intervals.
Table 5.5
Summary of Wind Tunnel Results for One Second Data Acquisition Periods
Time Period
(Seconds)
Measured AOA
Mean
(Degrees)
Corrected AOA
Mean
(Degrees)
Difference
Measured -Corrected
(Degrees)
16 to 17 -2.531 -2.473 -0.058
17 to 18 -2.513 -2.483 -0.030
18 to 19 -2.508 -2.478 -0.030
19 to 20 -2.550 -2.486 -0.064
20 to 21 -2.540 -2.481 -0.059
21 to 22 -2.511 -2.487 -0.024
97
Chapter 6
CONCLUDING REMARKS
An original system dynamic analysis approach is presented to evaluate the effects of
model vibrations on measured aerodynamic wind tunnel data. Analytical and
experimental results show that centrifugal accelerations associated with model vibration
cause bias errors in the inertial model attitude measurements. Wind-off dynamic response
tests on two transport model systems found bias errors over an order of magnitude greater
than the required device accuracy. An analysis is presented that shows these errors can
not be removed by filtering or averaging. Equations are developed to show the influence
of the model attitude errors on the determination of the drag coefficient.
A new time domain technique is developed to correct for the dynamically induced errors
in the inertial model attitude measurements using measured modal properties of the
model system. This modal technique extends previous work to compensate for multiple
modes in the pitch and yaw plane. Previously, the problem was associated with "sting
whip" with no detailed analysis of the underlying system dynamics. Dynamic response
tests on two transport models in a laboratory environment demonstrated the need to
compensate for multiple modes. Theoretical and experimental modal analyses are
presented to provide physical insight into the model system dynamics. Based on
observed rigid body model motion for the low frequency modes of interest, the problem is
simplified. For a planar rigid body model mode, analysis shows that the fuselage motion
can be completely described by a translation and rotation degree of freedom. A mode
98
radiusis definedto relatethetranslationandrotationdegreesof freedomusinganalytical
or experimentalmodeshapes.Analyses are presented that show the mode radii are not
affected significantly by the aerodynamic loads experienced in a high dynamic pressure
wind tunnel environment. A correlation coefficient is defined and used to validate the
rigid body model assumption.
Due to short data acquisition periods and the multi-mode random response observed in
wind tunnels, state of the art digital signal processing techniques are required to
implement the modal correction method in the time domain. Bandpass filters are used to
isolate the effects of individual modes and then the mode effects are combined using the
principle of superposition. During the filtering processes, the phase relationship of the
unfiltered model attitude signal and the model dynamic response must be maintained. To
achieve zero-phase distortion, finite impulse response filters are applied in both the
forward and reverse directions. The modal correction method compensates for the
dynamically induced bias error and provides a corrected model attitude time signal that
can be used to correlate with time varying changes in the balance forces
The modal correction method is verified through a series of wind-off dynamic response
tests and actual wind tunnel test data. The wind-off dynamic response tests show the
method has the ability to reduce the bias error in the inertial model attitude device by over
an order of magnitude to achieve the required device accuracy.
99
Theoreticalandexperimentalresultsarepresentedthat demonstratethe needto correct
for dynamicallyinducederrorsin inertialwind tunnelmodelattitudemeasurements.A
correctionmethodrequiringfour additionaltransducerswasdevelopedandimplemented
at the NationalAerospaceLaboratoryin the Netherlands.A principal advantageof the
modalcorrectiontechniqueis thatit minimizesthenumberof requiredtransducers(two)
usingthe modalpropertiesof the modelsystem. This is especiallycritical for models
with limited interior spaceandin wind tunnelsthathaveextremetemperatureconditions
where heated instrumentation packages are required. Recently redesigned
instrumentationpackagesfor the National TransonicFacility (NTF) provide the two
additionaltransducersrequiredfor themodalcorrectionmethod. Currently,facilities in
theUnitedStateshavenotimplementedacorrection.
Futureresearchof wind tunnel model systemdynamicsand its effectson measured
aerodynamicdata is recommendedin the following areas: (1) Perform a statistical
analysisto evaluatethe significanceof the magnitudeof the angleof attackcorrection
with respectto the measuredstandarddeviation,and small angleassumptionfor high
anglesof attack; (2) Performastudyof thecrossaxissensitivityof the inertial attitude
sensor,andtheeffectsof modelroll motions; (3) Performa studyof alternatesignal
processingmethods,such as modulationtechniques,for removing the dynamically
inducederrorsin the inertialmodelattitudemeasurements;(4) Basedon theobserved
rigid body modelbehavior,performa parametricstudyto evaluatechangesin dynamic
responsefor variationsin: massor massdistributionof themodel;balancestiffnessand
100
damping;and sting materialproperties. This researchwould be aimedat developing
designcriteriafor modelsystemsthatwouldminimize themodeldynamicresponseand
move closer to the desired steady-statewind tunnel test conditions. Further
enhancementsmaybe foundin theuseof activevibrationcontrol techniquesto suppress
themodelvibrations.
101
[1]
[2]
[31
[41
[51
[6]
[7]
[81
[9]
[10]
[11]
Chapter 7
REFERENCES
Young, C. P., Jr.:"Model Dynamics", AGARD Special Course on Cryogenic
Wind Tunnels, 1996.
Fuller, D.E.: "Guide to Users of the National Transonic Facility",
NASA TM-83124, July, 1981.
Strganac, T. W.: "A Study of the Aeroelastic Stability for the Model Support
System of the National Transonic Facility", AIAA-88-2033, 1988.
Whitlow, W., Jr.; Bennet, R. M.; and Strganac, T. W.: "Analysis of Vibrations of
the National Transonic Facility Model Support System Using a 3-D Aeroelastic
Code", AIAA-89-2207, 1989.
Young, C. P., Jr.; Popernack, T. G., Jr.; Gloss, B.B.: "National Transonic Facility
Model and Model Support Vibration Problems", AIAA-90-1416, 1990.
Buehrle, R. D.; Young, C. P., Jr.; Balakrishna, S.; and Kilgore, W. A.:
"Experimental Study of Dynamic Interaction Between Model Support Structure
and a High Speed Research Model in the National Transonic Facility",
AIAA-94-1623, 1994.
Young, C. P., Jr.; Buehrle, R. D.; Balakrishna, S.; and Kilgore, W.A.: "Effects ofVibration on Inertial Wind-Tunnel Model Attitude Measurement Devices".
NASA Technical Memorandum 109083, August, 1994.
Buehrle, R. D.; Young, C. P., Jr.; Burner, A. W.; Tripp, J. S.; Tcheng, P.; Finley,
T. D.; and Popernack, T. G., Jr.: "Dynamic Response Tests of Inertial and OpticalWind-Tunnel Model Attitude Measurement Devices", NASA Technical
Memorandum 109182, February, 1995.
Pope, A.; and Goin, K. L.: High Speed Wind Tunnel Testing, John Wiley &
Sons, Inc., New York, 1965.
Muhlstein,L. ,Jr.; and Coe, C. F.: "Integration Time Required to Extract Accurate
Data from Transonic Wind-Tunnel Tests", Journal of Aircraft, Volume 16, No. 9,
pp 620-625, September 1979.
Mabey, D. G.: "Flow Unsteadiness and Model Vibration in Wind Tunnels at
Subsonic and Transonic Speeds", Royal Aircraft Establishment Technical Report
70184, October, 1970.
102
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
Steinle, F. and Stanewsky, E.: "Wind Tunnel Flow Quality and Data Accuracy
Requirements", Advisory Group for Aerospace Research and Development
(AGARD) Advisory Report No. 184, November, 1982.
Finley, T., and Tcheng, P.: "Model Attitude Measurements at NASA Langley
Research Center", AIAA-92-0763, 1992.
Burt, G. E., and Uselton, J. C.: "Effect of Sting Oscillations on the Measurement
of Dynamic Stability Derivatives in Pitch and Yaw", AIAA Paper No. 74-612,
July, 1974.
Billingsley, J. P.: "Sting Dynamics of Wind Tunnel Models", Arnold Engineering
Development Center Report Number: AEDC-TR-76-41, May, 1976.
Fuijkschot, P. H.: "Use of Servo-Accelerometers for the Measurement of
Incidence of Windtunnel Models", National Aerospace Laboratory, The
Netherlands, Memorandum AW-84-008, 1984.
Buehrle, R. D.; and Young, C. P., Jr.; "Modal Correction Method for
Dynamically Induced Errors in Wind-Tunnel Model Attitude Measurements",
Proceedings of the 13th International Modal Analysis Conference, pp. 1708-1714,
Nashville, Tennessee, February 13-16, 1995.
Tcheng, P.; Tripp, J. S.; and Finley, T. D.; Effects of Yaw and Pitch Motion onModel Attitude Measurements, NASA Technical Memorandum 4641, February
1995.
Fuijkschot, P. H.: "A Correction Technique for Gravity Sensing Inclinometers",
National Aerospace Laboratory, The Netherlands, Memorandum AF-95-004, 1995
Fuijkschot, P. H.: "A Correction Technique for Gravity Sensing Inclinometers-
Phase 2: Proof of Concept", National Aerospace Laboratory, The Netherlands,
CR 95458L, 1995.
Gloss, Blair, B.; "Future Experimental Needs To Support Applied Aerodynamics:
A Transonic Perspective", AIAA Paper 92-0156, 1992.
Owen, F. K.; Orngard, G. M.; McDevitt, T. K.; and Ambur, T. A.; "A Dynamic
Optical Model Attitude Measurement System", European Transonic Windtunnel
GmbH and DFVLR, Cryogenic Technology Meeting, 2nd, Cologne, West
Germany, June 28-30, 1988, Paper, 21 p.
Roberson, J. A.; and Crowe, C. T.: Engineering Fluid Mechanics, Houghton
Mifflin Company, 1980.
103
[35]
[36]
[37]
[38]
[24] Meirovitch, Leonard: Elements of Vibration Analysis, McGraw-Hill, Inc., 1975,
p 240-250.
[25] Wells, D. A.: Schaum's Outline of Theory and Problems of Lagrangian
Dynamics, Schaum Publishing Company, 1967.
[26] Tse, F. S.; Morse, I. E.; and Hinkle, R. T.: Mechanical Vibrations Theory and
Applications, Allyn and Bacon, Inc., 1978.
[27] Fung, Y. C.: An Introduction to the Theory of Aeroelasticity,
John Wiley & Sons, Inc., 1955.
[28] Craig, R. R., Jr.: Structural Dynamics: An Introduction to Computer Methods,
John Wiley & Sons, Inc., 1981.
[29] Allemang, R. J.; and Brown, D.L.: Chapter 21: Experimental Modal Analysis,Shock and Vibration Handbook, 3rd Edition, McGraw Hill, Inc., 1988.
[30] Ewins, D. J.: Modal Testing: Theory and Practice, Research Studies Press LTD.,
1984
[31] Ferris, A.T.: "Cryogenic Wind Tunnel Force Instrumentation", NASA
Conference Publication No. 2122, Part II, 1982, pp 299-315.
[32] Thomson, W. T.: Vibration Theory and Applications, Prentice- Hall, Inc., 1965,
pp. 179-182.
[33] Davenport, A. G.; and Novak, M.: Chapter 29 Part II: Vibration of Structures
Induced by Wind, Shock and Vibration Handbook, 3rd Edition, McGraw Hill,
Inc., 1988.
[34] Alder, H. L.; and Roessler, E. B.: Introduction to Probability and Statistics,
W. H. Freeman and Company, 1960.
MATLAB Reference Guide, The Math Works Inc., August, 1992.
Hornbeck, R. W.; Numerical Methods, Quantum Publishers, Inc., 1975.
The STAR System User Manual, Spectral Dynamics, Inc., 1996.
MSC/NASTRAN User's Manual, The MacNeal-Schwendler Corporation, 1989
104
Appendix A
EFFECT OF AERODYNAMIC FORCES ON MODAL CHARACTERISTICS
Introduction
In this section, the effect of aerodynamic forces on the modal characteristics of a
cantilevered wind tunnel model system are examined. The objective is to validate the
assumption that the modal characteristics do not change significantly under the wind
tunnel test conditions. This is a fundamental assumption of the modal correction method
that enables wind-off estimates of the natural frequencies and mode effective radii to be
used for correction of the model attitude measurement during wind tunnel testing. A
finite element model (FEM) of a representative cantilevered transport model is used as
the basis for evaluating the modal characteristics for several loading conditions including
the most severe forces measured in a recent wind tunnel test on this model in the National
Transonic Facility (NTF).
Analytical Model
The finite element model of a representative cantilevered transport model system for the
NTF was generated and analyzed using the MSC/NASTRAN ® [38] structural analysis
program. The FEM was developed with the goal of representing the low frequency (less
than 50 Hertz) "rigid-fuselage" modes that contribute to the errors in the inertial model
attitude measurements. Detailed modeling of the wings was not of interest for this study.
The sting and model fuselage are constructed of beam elements with equivalent material
105
and cross-sectionspecificgeometricproperties. The forcebalancewhich connectsthe
stingto the fuselagewasmodeledusinga concentratedmassequalto thebalancemass
and rigid bar elements. Springswere usedat the connectionbetweenthe rigid bar
elementand the fuselageto representthe balancestiffnesscorrespondingto the three
translationandthreerotationdegreesof freedom.Thebalancestiffnesswasdetermined
from experimentalmeasurements.The wings are modeled as concentratedmasses
attachedto thefuselageusingrigid barelements. An additionallumpedmasswasused
to representinstrumentationandassociatedhardware.
The primary generalizedforcesare the unsteadyaerodynamicloads.
loads are modeled using a quasi-steadyapproximation [27].
aerodynamicforcesaremodeledas:
QF = q_ × S × CF
The aerodynamic
The generalized
(A.I)
where, q_ is the dynamic pressure and S is the characteristic area. The coefficient CF will
be assumed linear and is a function of the model attitude. Similarly, the generalized
aerodynamic moments are modeled as:
QM = qoo x S × d × C M (A.2)
where d is the characteristic length and the coefficient CM is assumed linear and is a
function of the model attitude.
Data from a high-speed (Mach =0.9, q==1800 pounds per square foot) wind tunnel test of
this transport model in the NTF were used to determine the four most severe loading
106
conditions. To add additional conservatism,this data was scaledup to a dynamic
pressureof 2700poundsper squarefoot. The resultingloadingconditionsare listed in
Table 1. Theseforcesandmomentswereappliedto theFEM at a point on the fuselage
coincidentwith thebalancemomentcenter.
Table 1Transport Model
Worst Case Loading Conditions
Load Case
1
Axial Force
(Pounds)
69
Normal Force
(Pounds)-2271
Pitch Moment
(Inch-Pounds)4OOO
2 63 -491 3158
3 -53 2688 1474
4 -184 6035 632
For each of the four different aerodynamic load cases, a static analysis was run to
generate a prestressed model and then the eigensolution was run for this prestressed
loading condition. The eigensolution was also run for the no load case to provide a
baseline set of natural frequencies and mode shapes.
Results and Conclusions
The purpose of the analysis was to assess the effect of aerodynamic loading on the modal
characteristics of a cantilevered wind tunnel model system. For the research presented in
this dissertation, an important constant is the modal radius which is estimated from a
linear regression fit of the fuselage mode shape coefficients. Therefore, the comparison
107
criteriaarenaturalfrequenciesandmoderadii.Themoderadiusfor thefirst six analytical
modeswereestimatedusingthemethoddescribedin Chapter4. The naturalfrequencies
and mode radii for the different loading conditions are listed in Tables 2 and 3,
respectively. The naturalfrequencydoesnot shift significantlyfor anyof the loading
conditions.For the largestaerodynamicforcesmeasuredon a representativetransport
model in the NationalTransonicFacility, the predictedshifts in the modal radiuswere
lessthan4%,which is negligible.
Table 2
Transport Model
Natural Frequency Comparison
No Load Load
Case 1
Mode Frequency Frequency
(Hz) (Hz)
1 9.19 9.19
2 9.23 9.23
3 17.2 17.2
4 17.3 17.4
5
Load Load Load
Case 2 Case 3 Case 4
Frequency Frequency Frequency Maximum
(Hz) (Hz) (Hz) Difference
(%)9.19 9.20 9.22 0.3
9.23 9.25 9.30 0.8
17.2 17.2 17.3 0.6
17.4 17.4 17.4 0.6
29.5 29.6 29.7 0.729.5 29.5
30.4 30.46 30.4 30.5 30.6 0.7
Note: * Difference (%) = (fioaa-fnoload '/fnoload * 100
108
Table 3
Transport Model
Mode Radius Comparison
Mode
No Load
Radius
(Inch)
Load
Case 1
Radius
(Inch)
Load
Case 2
Radius
(Inch)
Load
Case 3
Radius
(Inch)
Load
Case 4
Radius
(Inch)
Maximum
Difference
(%)*39.4
Note:
39.7
39.639.5 39.4
39.8 39.7
7.68 7.67
8.24 8.20
-3.66 -3.67
-3.18 -3.17
40.0
40.2
41.1
1.8
3.5
3 7.68 7.72 7.87 2.5
4 8.21 8.28 8.54 4.0
5 -3.67 -3.67 -3.64 -0.8
6 -3.17 -3.15 -3.14 -0.9
* Difference (%) = (Rload-R.olo_)/R.olo.d * 100
109